lav2

LAV2 package root.

Modules:

Name	Description
assets	Asset loading helpers for LAV2.
controller	Controller package for LAV2.
dynamics	Dynamics models for LAV2.
gamepads	Gamepad input helpers for LAV2.
runner	Runner utilities for LAV2 training and evaluation.
tasks	Task packages for supported training frameworks for LAV2.
trajectories	Trajectory implementations package for LAV2.

assets

Asset loading helpers for LAV2.

controller

Controller package for LAV2.

Modules:

Name	Description
__main__	Entry point for running the LAV2 controller demo.
base	Base interfaces for LAV2 controllers.
geo	Geometric controllers and helpers for SE(3) tracking.
mapping	Control-mode mapping utilities for NumPy-based controller pipelines.
mixer	Command mixer for mapping thrust/moment demands to rotor speeds.
pid	PID controllers for flight and track control.
run	Interactive MuJoCo runner for flight and track controllers.
torch	Torch-based controller implementations.
utils	Logging utilities for controller runs.
xadap	Python implementation of Learning-based Controller for Extreme Adaptation.

__main__

Entry point for running the LAV2 controller demo.

base

Base interfaces for LAV2 controllers.

Classes:

Name	Description
ControllerBase	Abstract controller base class.

ControllerBase

ControllerBase(params: VehicleParams, control_mask: ndarray | Tensor, gains: dict | None = None)

Bases: ABC

Abstract controller base class.

Initialize shared controller state.

Methods:

Name	Description
update	Update the controller and return control outputs.
reset	Reset controller state if necessary.

Attributes:

Name	Type	Description
params	VehicleParams	Vehicle parameters for the controller.
control_mask	ndarray | Tensor	unified control mask for target/state channels, used in update(). e.g. using 9~11 for attitude yaw control.
gains	dict | None	Optional gains configuration for derived controllers.
target	ndarray | Tensor	Latest target input to the controller.

params instance-attribute

params: VehicleParams = params

Vehicle parameters for the controller.

control_mask instance-attribute

control_mask: ndarray | Tensor = control_mask

unified control mask for target/state channels, used in update(). e.g. using 9~11 for attitude yaw control.

gains instance-attribute

gains: dict | None = gains

Optional gains configuration for derived controllers.

target instance-attribute

target: ndarray | Tensor

Latest target input to the controller.

update abstractmethod

update(target, state)

Update the controller and return control outputs.

reset abstractmethod

reset(*args, **kwargs)

Reset controller state if necessary.

geo

Geometric controllers and helpers for SE(3) tracking.

Classes:

Name	Description
GeoControl	Python implementation of Geometric Controller.
GeometricAdaptiveController	Python implementation of Geometric Adaptive Controller.
L1_GeoControl	Python implementation of L1 Adaptive Geometric Controller.

Functions:

Name	Description
euler_zyx_to_R	Convert ZYX Euler angles into a rotation matrix.

GeoControl

GeoControl(params: VehicleParams = VehicleParams(), control_mask: list[int] = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], gains: dict | None = None)

Bases: ControllerBase

Python implementation of Geometric Controller.

Based on: Taeyoung Lee, Melvin Leok, and N.H. McClamroch. "Geometric Tracking Control of a Quadrotor UAV on SE(3)" [DOI: 10.1109/CDC.2010.5717652].

Initialize gains and error buffers.

Parameters:

Name	Type	Description	Default
params	VehicleParams	Vehicle parameters.	VehicleParams()
control_mask	list[int]	Control mask for target/state channels.	[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
gains	dict | None	Optional geometric gains for keys x/v/R/W.	None

Methods:

Name	Description
reset	Reset controller state.
update	Compute thrust and moments from target and state.

Attributes:

Name	Type	Description
params	VehicleParams	Vehicle parameters for the controller.
control_mask	ndarray | Tensor	unified control mask for target/state channels, used in update(). e.g. using 9~11 for attitude yaw control.

params instance-attribute

params: VehicleParams = params

Vehicle parameters for the controller.

control_mask instance-attribute

control_mask: ndarray | Tensor = control_mask

unified control mask for target/state channels, used in update(). e.g. using 9~11 for attitude yaw control.

reset

reset()

Reset controller state.

update

update(target: ndarray, state: ndarray) -> np.ndarray

Compute thrust and moments from target and state.

Parameters:

Name	Type	Description	Default
target	ndarray	Desired state vector.	required
state	ndarray	Current state vector.	required

Returns:

Type	Description
ndarray	np.ndarray: Control output ordered as [thrust, Mx, My, Mz].

GeometricAdaptiveController

GeometricAdaptiveController(params: VehicleParams = VehicleParams(), control_mask: list[int] = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], gains: dict | None = None)

Bases: GeoControl

Python implementation of Geometric Adaptive Controller.

Based on: Farhad A. Goodarzi, Daewon Lee, and Taeyoung Lee. "Geometric Adaptive Tracking Control of a Quadrotor Unmanned Aerial Vehicle on SE(3) for Agile Maneuvers." [DOI: 10.1115/1.4030419].

Initialize adaptive gains and inherited geometry control.

Parameters:

Name	Type	Description	Default
params	VehicleParams	Vehicle parameters.	VehicleParams()
control_mask	list[int]	Control mask for target/state channels.	[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
gains	dict | None	Optional geometric gains for keys x/v/R/W.	None

Methods:

Name	Description
update	Adaptive geometric control using target/state vectors (GeoControl layout).
reset	Reset controller state.

Attributes:

Name	Type	Description
params	VehicleParams	Vehicle parameters for the controller.
control_mask	ndarray | Tensor	unified control mask for target/state channels, used in update(). e.g. using 9~11 for attitude yaw control.

params instance-attribute

params: VehicleParams = params

Vehicle parameters for the controller.

control_mask instance-attribute

control_mask: ndarray | Tensor = control_mask

unified control mask for target/state channels, used in update(). e.g. using 9~11 for attitude yaw control.

update

update(target: ndarray, state: ndarray) -> np.ndarray

Adaptive geometric control using target/state vectors (GeoControl layout).

Returns:

Type	Description
ndarray	np.ndarray: Control output ordered as [thrust, Mx, My, Mz].

reset

reset()

Reset controller state.

L1_GeoControl

L1_GeoControl(params: VehicleParams = VehicleParams(), control_mask: list[int] = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], gains: dict | None = None)

Bases: GeoControl

Python implementation of L1 Adaptive Geometric Controller.

Based on: Zhuohuan Wu, Sheng Cheng, Pan Zhao, Aditya Gahlawat, Kasey A. Ackerman, Arun Lakshmanan, Chengyu Yang, Jiahao Yu, Naira Hovakimyan. "L1Quad: L1 Adaptive Augmentation of Geometric Control for Agile Quadrotors with Performance Guarantees" [arXiv:2302.07208].

Initialize L1 adaptive parameters.

Parameters:

Name	Type	Description	Default
params	VehicleParams	Vehicle parameters.	VehicleParams()
control_mask	list[int]	Control mask for target/state channels.	[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
gains	dict | None	Optional geometric gains for keys x/v/R/W.	None

Methods:

Name	Description
update	L1-augmented geometric control using target/state vectors.
reset	Reset controller state.

Attributes:

Name	Type	Description
dt_L1		For large uncertainties ...
L1_params		Initialization of L1 inputs
params	VehicleParams	Vehicle parameters for the controller.
control_mask	ndarray | Tensor	unified control mask for target/state channels, used in update(). e.g. using 9~11 for attitude yaw control.

dt_L1 instance-attribute

dt_L1 = sim_dt

For large uncertainties ...

L1_params instance-attribute

L1_params = (As_v, As_omega, dt_L1, ctoffq1Thrust, ctoffq1Moment, ctoffq2Moment, mass, g, inertia)

Initialization of L1 inputs

params instance-attribute

params: VehicleParams = params

Vehicle parameters for the controller.

control_mask instance-attribute

control_mask: ndarray | Tensor = control_mask

unified control mask for target/state channels, used in update(). e.g. using 9~11 for attitude yaw control.

update

update(target: ndarray, state: ndarray) -> np.ndarray

L1-augmented geometric control using target/state vectors.

Parameters:

Name	Type	Description	Default
target	ndarray	12-dim vector [x,y,z,vx,vy,vz,roll,pitch,yaw,wx,wy,wz].	required
state	ndarray	12-dim vector [x,y,z,vx,vy,vz,roll,pitch,yaw,wx,wy,wz].	required

Returns:

Type	Description
ndarray	np.ndarray: Control output ordered as [thrust, Mx, My, Mz].

reset

reset()

Reset controller state.

euler_zyx_to_R

euler_zyx_to_R(yaw: float, pitch: float, roll: float) -> np.ndarray

Convert ZYX Euler angles into a rotation matrix.

Returns:

Type	Description
ndarray	np.ndarray: 3x3 rotation matrix.

mapping

Control-mode mapping utilities for NumPy-based controller pipelines.

Classes:

Name	Description
FlightMappingLimits	Per-channel limits used when mapping normalized actions to targets.
FlightMappingResult	Mapped result for a normalized flight action or action batch.
FlightActionMapper	Map normalized actions into controller targets, wrench commands, or RPM.

Functions:

Name	Description
target_to_command	Convert a full 12D target into a compact command vector.

FlightMappingLimits dataclass

FlightMappingLimits(vel_limits: tuple[float, float, float] = (5.0, 5.0, 5.0), att_limits: tuple[float, float, float] = (1.0, 1.0, math.pi), br_limits: tuple[float, float, float] = (2.0, 2.0, 2.0))

Per-channel limits used when mapping normalized actions to targets.

FlightMappingResult dataclass

FlightMappingResult(clamped_action: ndarray | Tensor, rpm_command: ndarray | Tensor, control: ndarray | Tensor | None = None, target: ndarray | Tensor | None = None)

Mapped result for a normalized flight action or action batch.

FlightActionMapper

FlightActionMapper(control_mode: str, params: VehicleParams = VehicleParams(), limits: FlightMappingLimits = FlightMappingLimits())

Map normalized actions into controller targets, wrench commands, or RPM.

Initialize a NumPy flight action mapper for the selected control mode.

Methods:

Name	Description
map_action	Map a normalized action into an RPM command and optional intermediates.

Attributes:

Name	Type	Description
requires_controller	bool	Whether this mode needs a controller target expansion step.

requires_controller property

requires_controller: bool

Whether this mode needs a controller target expansion step.

map_action

map_action(action: ndarray, mixer: Mixer, *, state: ndarray | None = None, flight_controller: ControllerBase | None = None) -> FlightMappingResult

Map a normalized action into an RPM command and optional intermediates.

target_to_command

target_to_command(target: ndarray, command_mode: str, state: ndarray | None = None) -> np.ndarray

Convert a full 12D target into a compact command vector.

Parameters:

Name	Type	Description	Default
target	ndarray	Full target state ordered as [x, y, z, vx, vy, vz, roll, pitch, yaw, wx, wy, wz].	required
command_mode	str	Supported target command mode. Valid values are "tgt_pos", "tgt_vel", "tgt_ctatt", and "tgt_ctbr".	required
state	ndarray | None	Current 12D vehicle state. Required only for "tgt_pos", where the returned command is the body-frame position error and heading error rather than an absolute world-frame pose target.	None

Returns:

Name	Type	Description
	ndarray	np.ndarray: A compact command vector matching the requested command
mode	ndarray	tgt_pos -> [dx_body, dy_body, dz, dyaw] tgt_vel -> [vx, vy, vz, yaw_rate] tgt_ctatt -> [collective, roll, pitch, yaw_rate] tgt_ctbr -> [collective, wx, wy, wz]

Raises:

Type	Description
ValueError	If command_mode is unsupported or state is missing for tgt_pos.

mixer

Command mixer for mapping thrust/moment demands to rotor speeds.

Classes:

Name	Description
Mixer	Map thrust and moment demands to rotor commands.

Mixer

Mixer(params: VehicleParams = VehicleParams(), verbose=False)

Map thrust and moment demands to rotor commands.

Note

电机推力分配： 电机1 = 推力 - 横滚(roll) + 俯仰(pitch) + 偏航(yaw) 电机2 = 推力 + 横滚(roll) - 俯仰(pitch) + 偏航(yaw) 电机3 = 推力 + 横滚(roll) + 俯仰(pitch) - 偏航(yaw) 电机4 = 推力 - 横滚(roll) - 俯仰(pitch) - 偏航(yaw)

Initialize the mixer with vehicle parameters and mixer matrix.

Methods:

Name	Description
apply_thrust_curve	Apply thrust curve to normalized thrust command.
calculate_rotor_commands	Allocate thrust and moments into rotor speeds.

Attributes:

Name	Type	Description
mat		转速平方到推力扭矩
inv_mat		推力扭矩到转速平方
max_rpm_sq		最大转速平方
min_rpm_sq		最小转速平方
max_wrench		单桨最大推力/力矩
min_wrench		单桨最小推力/力矩
max_collective_thrust		最大总推力

mat instance-attribute

mat = array([[cT * cos(alpha)] * 4, [cT * cos(alpha) * y] * 4, [cT * cos(alpha) * x] * 4, [cT * sin(alpha) * x + cM] * 4]) * array([[1, 1, 1, 1], [-1, 1, 1, -1], [-1, 1, -1, 1], [1, 1, -1, -1]])

转速平方到推力扭矩

inv_mat instance-attribute

inv_mat = pinv(mat)

推力扭矩到转速平方

max_rpm_sq instance-attribute

max_rpm_sq = max_rpm ** 2

最大转速平方

min_rpm_sq instance-attribute

min_rpm_sq = init_rpm ** 2

最小转速平方

max_wrench instance-attribute

max_wrench = mat[:, 1] * max_rpm_sq

单桨最大推力/力矩

min_wrench instance-attribute

min_wrench = mat[:, 1] * min_rpm_sq

单桨最小推力/力矩

max_collective_thrust instance-attribute

max_collective_thrust = max_wrench[0] * 4.0

最大总推力

apply_thrust_curve

apply_thrust_curve(normalized_thrust: float, mode: int = 1) -> float

Apply thrust curve to normalized thrust command.

References

• cmd_ctatt: https://github.com/PX4/PX4-Autopilot/blob/e5a483e4b71ab4b994cb0121fe28992d8ce0c8cd/src/modules/mc_att_control/mc_att_control_main.cpp#L111-L137
• cmd_ctbr: https://github.com/PX4/PX4-Autopilot/blob/e5a483e4b71ab4b994cb0121fe28992d8ce0c8cd/src/modules/mc_rate_control/MulticopterRateControl.cpp#L156-L179

Parameters:

Name	Type	Description	Default
normalized_thrust	float	Normalized thrust command [-1, 1].	required
mode	int	Thrust curve mode.	1

Returns:

Name	Type	Description
float	float	Total thrust in Newtons.

Raises:

Type	Description
NotImplementedError	If the specified mode is not implemented.

calculate_rotor_commands

calculate_rotor_commands(control: ndarray) -> np.ndarray

Allocate thrust and moments into rotor speeds.

Parameters:

Name	Type	Description	Default
control	ndarray	Desired total thrust and roll/pitch/yaw moments. Shape: (4,).	required

Returns:

Type	Description
ndarray	np.ndarray: The amplitude of each motor's rotational speed (calculated using sqrt(clip(rpm_sq))). Shape: (4,).

pid

PID controllers for flight and track control.

Classes:

Name	Description
PID	Classic PID controller with optional derivative filtering.
FlightController	PID-based flight controller.
TrackController	PID-based track controller.

PID

PID(Kp: Any = 0.0, Ki: Any = 0.0, Kd: Any = 0.0, dt: float = 0.01, N: float = 100.0, integral_limit: Any | None = None)

Classic PID controller with optional derivative filtering.

The internal state shape follows Kp through NumPy broadcasting. This makes scalar and multi-axis gains behave consistently without additional shape bookkeeping.

Initialize PID gains and internal state.

Parameters:

Name	Type	Description	Default
Kp	Any	Proportional gain. Its shape defines the PID state shape.	0.0
Ki	Any	Integral gain broadcast against the PID state.	0.0
Kd	Any	Derivative gain broadcast against the PID state.	0.0
dt	float	Controller time step.	0.01
N	float	First-order derivative filter coefficient.	100.0
integral_limit	Any | None	Optional elementwise clamp for the integral state.	None

Methods:

Name	Description
reset	Reset integrator and derivative state.
update	Compute the PID output for the current error.

reset

reset() -> None

Reset integrator and derivative state.

update

update(error: ndarray | float) -> np.ndarray

Compute the PID output for the current error.

Parameters:

Name	Type	Description	Default
error	ndarray | float	Control error compatible with the PID state shape derived from Kp.	required

Returns:

Type	Description
ndarray	np.ndarray: Combined P+I+D control output.

FlightController

FlightController(params: VehicleParams = VehicleParams(), control_mask: list[int] = [1, 1, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0], gains: dict | None = None, limits: dict | None = None)

Bases: ControllerBase

PID-based flight controller.

Initialize the flight PID controller.

Parameters:

Name	Type	Description	Default
params	VehicleParams	Vehicle parameters used by the controller.	VehicleParams()
control_mask	list[int]	Boolean-like mask indicating which target channels are provided externally and which are filled by the controller.	[1, 1, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0]
gains	dict | None	Optional PID gain dictionary. If omitted, the module-level flight defaults are used.	None
limits	dict | None	Optional command-limit configuration. Supported keys include vel_cmd, vel_error and att_cmd. Any key can be set to None to disable that clamp.	None

Methods:

Name	Description
reset	Reset all inner PID loops of the flight controller.
update	Compute thrust and moments for the current target/state.

Attributes:

Name	Type	Description
params	VehicleParams	Vehicle parameters for the controller.
control_mask	ndarray | Tensor	unified control mask for target/state channels, used in update(). e.g. using 9~11 for attitude yaw control.

params instance-attribute

params: VehicleParams = params

Vehicle parameters for the controller.

control_mask instance-attribute

control_mask: ndarray | Tensor = control_mask

unified control mask for target/state channels, used in update(). e.g. using 9~11 for attitude yaw control.

reset

reset() -> None

Reset all inner PID loops of the flight controller.

update

update(target: ndarray, state: ndarray) -> np.ndarray

Compute thrust and moments for the current target/state.

Parameters:

Name	Type	Description	Default
target	ndarray	Desired state vector with shape (12,) ordered as [x, y, z, vx, vy, vz, roll, pitch, yaw, wx, wy, wz].	required
state	ndarray	Current state vector with the same ordering as target.	required

Returns:

Type	Description
ndarray	np.ndarray: Control output [thrust, roll_moment, pitch_moment, yaw_moment].

TrackController

TrackController(params: VehicleParams = VehicleParams(), control_mask: list[int] = [0, 0, 0, 1, 0, 1], gains: dict | None = None)

Bases: ControllerBase

PID-based track controller.

Initialize the track PID controller.

Parameters:

Name	Type	Description	Default
params	VehicleParams	Vehicle parameters used by the controller.	VehicleParams()
control_mask	list[int]	Boolean-like mask indicating which target channels are provided externally and which are filled by the controller.	[0, 0, 0, 1, 0, 1]
gains	dict | None	Optional PID gain dictionary. If omitted, the module-level track defaults are used.	None

Methods:

Name	Description
reset	Reset all inner PID loops of the track controller.
update	Compute track accelerations for the current target/state.

Attributes:

Name	Type	Description
params	VehicleParams	Vehicle parameters for the controller.
control_mask	ndarray | Tensor	unified control mask for target/state channels, used in update(). e.g. using 9~11 for attitude yaw control.

params instance-attribute

params: VehicleParams = params

Vehicle parameters for the controller.

control_mask instance-attribute

control_mask: ndarray | Tensor = control_mask

unified control mask for target/state channels, used in update(). e.g. using 9~11 for attitude yaw control.

reset

reset() -> None

Reset all inner PID loops of the track controller.

update

update(target: ndarray, state: ndarray) -> np.ndarray

Compute track accelerations for the current target/state.

Parameters:

Name	Type	Description	Default
target	ndarray	Desired state vector with shape (6,) ordered as [x, y, yaw, u, v, r].	required
state	ndarray	Current state vector with the same ordering as target.	required

Returns:

Type	Description
ndarray	np.ndarray: Left and right track acceleration commands [a_l, a_r].

run

Interactive MuJoCo runner for flight and track controllers.

Functions:

Name	Description
key_callback	Handle keyboard target updates.
load_callback	Load the MuJoCo model and data for the viewer.
control_callback	Run the control loop for the active mode.
main	Launch the interactive demo with CLI options.

key_callback

key_callback(keycode: int) -> None

Handle keyboard target updates.

load_callback

load_callback() -> tuple[mujoco.MjModel, mujoco.MjData]

Load the MuJoCo model and data for the viewer.

Returns:

Name	Type	Description
tuple	tuple[MjModel, MjData]	(m, d) MuJoCo model and data objects.

control_callback

control_callback(m: MjModel, d: MjData) -> None

Run the control loop for the active mode.

main

main(argv: list[str] | None = None) -> None

Launch the interactive demo with CLI options.

Parameters:

Name	Type	Description	Default
argv	list[str] | None	Optional argument list to parse.	None

Notes

--mode: 0=flight (default), 1=track. --control: Flight controller: pid|geo (default pid). Track: pid. --control-mode: Flight control mode (cmd_pos|cmd_vel|cmd_ctatt|cmd_ctbr). --gamepad: Enable gamepad input. --trajectory: Enable automatic trajectory tracking for flight cmd_pos/cmd_vel only.

torch

Torch-based controller implementations.

Modules:

Name	Description
mapping	Control-mode mapping utilities for batched Torch controller pipelines.
mixer	Torch mixer for mapping thrust/moment demands to rotor speeds.
pid	Torch-based PID controllers for batched simulation.

mapping

Control-mode mapping utilities for batched Torch controller pipelines.

Classes:

Name	Description
FlightActionMapper	Map batched normalized actions into controller targets, wrench, or RPM.

Functions:

Name	Description
target_to_command	Convert full 12D target batches into compact command vectors.

FlightActionMapper

FlightActionMapper(control_mode: str, params: VehicleParams = VehicleParams(), limits: FlightMappingLimits = FlightMappingLimits(), *, num_envs: int = 1, device: str | device = 'cpu')

Map batched normalized actions into controller targets, wrench, or RPM.

Initialize a batched Torch flight action mapper for the selected mode.

Methods:

Name	Description
map_action	Map a normalized batched action into RPM commands and intermediates.

Attributes:

Name	Type	Description
requires_controller	bool	Whether this mode needs a controller target expansion step.

requires_controller property

requires_controller: bool

Whether this mode needs a controller target expansion step.

map_action

map_action(action: Tensor, mixer: Mixer, *, state: Tensor | None = None, flight_controller: ControllerBase | None = None) -> FlightMappingResult

Map a normalized batched action into RPM commands and intermediates.

target_to_command

target_to_command(target: Tensor, command_mode: str, state: Tensor | None = None) -> torch.Tensor

Convert full 12D target batches into compact command vectors.

Parameters:

Name	Type	Description	Default
target	Tensor	Batched target states with shape (N, 12) ordered as [x, y, z, vx, vy, vz, roll, pitch, yaw, wx, wy, wz].	required
command_mode	str	Supported target command mode. Valid values are "tgt_pos", "tgt_vel", "tgt_ctatt", and "tgt_ctbr".	required
state	Tensor | None	Current batched 12D vehicle state with shape (N, 12). Required only for "tgt_pos", where the returned command is the body-frame position error and heading error.	None

Returns:

Type	Description
Tensor	torch.Tensor: A batched compact command tensor with shape (N, 4)
Tensor	matching the requested mode: - tgt_pos -> [dx_body, dy_body, dz, dyaw] - tgt_vel -> [vx, vy, vz, yaw_rate] - tgt_ctatt -> [collective, roll, pitch, yaw_rate] - tgt_ctbr -> [collective, wx, wy, wz]

Raises:

Type	Description
ValueError	If command_mode is unsupported or state is missing for tgt_pos.

mixer

Torch mixer for mapping thrust/moment demands to rotor speeds.

Classes:

Name	Description
Mixer	Map thrust and moment demands to rotor commands.

Mixer

Mixer(params: VehicleParams = VehicleParams(), num_envs: int = 1, device: str | device = 'cpu')

Map thrust and moment demands to rotor commands.

Note

电机推力分配： 电机1 = 推力 - 横滚(roll) + 俯仰(pitch) + 偏航(yaw) 电机2 = 推力 + 横滚(roll) - 俯仰(pitch) + 偏航(yaw) 电机3 = 推力 + 横滚(roll) + 俯仰(pitch) - 偏航(yaw) 电机4 = 推力 - 横滚(roll) - 俯仰(pitch) - 偏航(yaw)

Initialize the torch mixer with vehicle parameters and device.

Methods:

Name	Description
randomize	Randomize mixer runtime parameters from nominal values.
apply_thrust_curve	Apply thrust curve to normalized thrust command.
calculate_rotor_commands	Allocate thrust and moments into rotor speeds.

randomize

randomize(env_ids=None, randomization=None)

Randomize mixer runtime parameters from nominal values.

apply_thrust_curve

apply_thrust_curve(normalized_thrust: Tensor, mode: int = 1) -> torch.Tensor

Apply thrust curve to normalized thrust command.

References

cmd_ctatt: https://github.com/PX4/PX4-Autopilot/blob/e5a483e4b71ab4b994cb0121fe28992d8ce0c8cd/src/modules/mc_att_control/mc_att_control_main.cpp#L111-L137 cmd_ctbr: https://github.com/PX4/PX4-Autopilot/blob/e5a483e4b71ab4b994cb0121fe28992d8ce0c8cd/src/modules/mc_rate_control/MulticopterRateControl.cpp#L156-L179

Parameters:

Name	Type	Description	Default
normalized_thrust	Tensor	Normalized thrust command [-1, 1]. Shape: arbitrary, e.g. (num_envs, n_motors). Applied elementwise.	required
mode	int	Thrust curve mode.	1

Returns:

Type	Description
Tensor	torch.Tensor: Thrust in Newtons with the same shape as normalized_thrust.

Raises:

Type	Description
NotImplementedError	If the specified mode is not implemented.

calculate_rotor_commands

calculate_rotor_commands(control: Tensor) -> torch.Tensor

Allocate thrust and moments into rotor speeds.

Parameters:

Name	Type	Description	Default
control	Tensor	Desired total thrust and roll/pitch/yaw moments. Shape: (num_envs, 4)	required

Returns:

Type	Description
Tensor	torch.Tensor: The amplitude of each motor's rotational speed (calculated using sqrt(clip(rpm_sq))). Shape: (num_envs, 4)

pid

Torch-based PID controllers for batched simulation.

Classes:

Name	Description
PID	Vectorized PID controller for multiple environments.
FlightController	PID flight controller (batched).
TrackController	PID track controller (batched).

PID

PID(Kp: Tensor, Ki: Tensor | None = None, Kd: Tensor | None = None, dt: float = 0.01, N: float = 100.0, integral_limit: Tensor | None = None, num_envs: int = 1)

Vectorized PID controller for multiple environments.

Initialize PID gains and state tensors.

Parameters:

Name	Type	Description	Default
Kp	Tensor	Proportional gain tensor with a leading environment dimension.	required
Ki	Tensor | None	Integral gain tensor with the same shape as Kp.	None
Kd	Tensor | None	Derivative gain tensor with the same shape as Kp.	None
dt	float	Controller time step.	0.01
N	float	First-order derivative filter coefficient.	100.0
integral_limit	Tensor | None	Optional clamp applied elementwise to the integral state.	None
num_envs	int	Number of environments.	1

Methods:

Name	Description
reset	Reset PID state for selected environments.
update	Compute PID output for the given error tensor.

reset

reset(env_ids: slice | Sequence[int] | Tensor) -> None

Reset PID state for selected environments.

Parameters:

Name	Type	Description	Default
env_ids	slice | Sequence[int] | Tensor	Indices or slice selecting which environments to reset.	required

update

update(error: Tensor) -> torch.Tensor

Compute PID output for the given error tensor.

Parameters:

Name	Type	Description	Default
error	Tensor	Batched control error tensor compatible with the internal state shape determined from Kp.	required

Returns:

Type	Description
Tensor	torch.Tensor: Combined P+I+D control output.

FlightController

FlightController(params: VehicleParams = VehicleParams(), control_mask: list[int] = [1, 1, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0], gains: dict | None = None, limits: dict | None = None, num_envs: int = 1, device: str | device = 'cpu')

Bases: ControllerBase

PID flight controller (batched).

Initialize the batched flight PID controller.

Methods:

Name	Description
randomize	Randomize configured PID gains and command limits for selected environments.
reset	Reset PID loops for selected environments.
update	Update controller.

Attributes:

Name	Type	Description
params	VehicleParams	Vehicle parameters for the controller.
control_mask	ndarray | Tensor	unified control mask for target/state channels, used in update(). e.g. using 9~11 for attitude yaw control.

params instance-attribute

params: VehicleParams = params

Vehicle parameters for the controller.

control_mask instance-attribute

control_mask: ndarray | Tensor = control_mask

unified control mask for target/state channels, used in update(). e.g. using 9~11 for attitude yaw control.

randomize

randomize(env_ids=None, gains=None, limits=None)

Randomize configured PID gains and command limits for selected environments.

reset

reset(env_ids)

Reset PID loops for selected environments.

update

update(target: Tensor, state: Tensor) -> torch.Tensor

Update controller.

Parameters:

Name	Type	Description	Default
target	Tensor	(N, 12) desired [x, y, z, vx, vy, vz, roll, pitch, yaw, wx, wy, wz]	required
state	Tensor	(N, 12) current state same ordering.	required

Returns:

Type	Description
Tensor	torch.Tensor: Tensor of shape (N, 4): [thrust, roll_moment, pitch_moment, yaw_moment].

TrackController

TrackController(params: VehicleParams = VehicleParams(), control_mask: list[int] = [0, 0, 0, 1, 0, 1], gains: dict | None = None, num_envs: int = 1, device: str | device = 'cpu')

Bases: ControllerBase

PID track controller (batched).

Initialize the batched track controller.

Methods:

Name	Description
randomize	Randomize configured PID gains for all or selected environments.
reset	Reset PID loops for selected environments.
update	Update controller.

Attributes:

Name	Type	Description
params	VehicleParams	Vehicle parameters for the controller.
control_mask	ndarray | Tensor	unified control mask for target/state channels, used in update(). e.g. using 9~11 for attitude yaw control.

params instance-attribute

params: VehicleParams = params

Vehicle parameters for the controller.

control_mask instance-attribute

control_mask: ndarray | Tensor = control_mask

unified control mask for target/state channels, used in update(). e.g. using 9~11 for attitude yaw control.

randomize

randomize(env_ids=None, gains=None)

Randomize configured PID gains for all or selected environments.

reset

reset(env_ids)

Reset PID loops for selected environments.

update

update(target: Tensor, state: Tensor) -> torch.Tensor

Update controller.

Parameters:

Name	Type	Description	Default
target	Tensor	(N, 6) desired [x, y, yaw, u, v, r]	required
state	Tensor	(N, 6) current state same ordering.	required

Returns:

Type	Description
Tensor	torch.Tensor: (N, 2) left/right acceleration commands.

utils

Logging utilities for controller runs.

Classes:

Name	Description
Logger	A class for logging and visualization.

Logger

Logger(logging_freq_hz: int, output_folder: str = 'results', num_drones: int = 1, duration_sec: int = 0, colab: bool = False)

A class for logging and visualization.

Stores, saves to file, and plots the kinematic information and RPMs of a simulation with one or more drones.

Initialize logger buffers and output settings.

Note

The order in which information is stored by Logger.log() is not the same as the one in obs["id"]["state"]; see implementation below.

Parameters:

Name	Type	Description	Default
logging_freq_hz	int	Logging frequency in Hz.	required
output_folder	str	Output directory for saved logs.	'results'
num_drones	int	Number of drones.	1
duration_sec	int	Preallocated duration in seconds (0 disables).	0
colab	bool	Enable Colab-friendly plotting.	False

Methods:

Name	Description
log	Log entries for a single simulation step.
save	Save the logs to file.
save_as_csv	Save logs as comma separated values.
plot	Plot kinematics and RPMs for all drones.

log

log(drone: int, timestamp: float, state: ndarray, control: ndarray = np.zeros(12))

Log entries for a single simulation step.

Parameters:

Name	Type	Description	Default
drone	int	Drone index for the log entry.	required
timestamp	float	Simulation timestamp.	required
state	ndarray	Drone state vector, shape (20,).	required
control	ndarray	Control target vector, shape (12,).	zeros(12)

save

save()

Save the logs to file.

save_as_csv

save_as_csv(comment: str = '')

Save logs as comma separated values.

Parameters:

Name	Type	Description	Default
comment	str	Added to the folder name.	''

plot

plot()

Plot kinematics and RPMs for all drones.

xadap

Python implementation of Learning-based Controller for Extreme Adaptation.

Based on: Dingqi Zhang, Antonio Loquercio, Jerry Tang, Ting-Hao Wang, Jitendra Malik, Mark W. Mueller. "A Learning-based Quadcopter Controller with Extreme Adaptation" IEEE Transactions on Robotics, 41(2):3948-3964, 2025. https://doi.org/10.1109/TRO.2025.3577037

Classes:

Name	Description
ModelType	Enumeration for model normalization variants.
ButterworthFilter	Low-pass filter for smoothing sensor measurements.
NeuralNetworkModel	Neural network model for adaptation.
Xadap_NN_control	Controller combining high-level controller with neural adaptation.

ModelType

Bases: Enum

Enumeration for model normalization variants.

ButterworthFilter

ButterworthFilter(cutoff_freq=30, sample_rate=100)

Low-pass filter for smoothing sensor measurements.

Initialize filter coefficients and buffers.

Parameters:

Name	Type	Description	Default
cutoff_freq	float	Cutoff frequency in Hz.	30
sample_rate	float	Sampling rate in Hz.	100

Methods:

Name	Description
run	Filter a new sample and return the smoothed value.

run

run(x_new: ndarray) -> np.ndarray

Filter a new sample and return the smoothed value.

Parameters:

Name	Type	Description	Default
x_new	ndarray	New sample vector.	required

Returns:

Type	Description
ndarray	np.ndarray: Filtered sample.

NeuralNetworkModel

NeuralNetworkModel()

Neural network model for adaptation.

Initialize ONNX model paths, load normalization parameters, and activate sessions.

Methods:

Name	Description
activate	Initialize ONNX runtime sessions.
normalize_obs	Normalize observations based on model type.
predict	Run inference through both networks.

activate

activate()

Initialize ONNX runtime sessions.

normalize_obs

normalize_obs(obs: ndarray, model_type: ModelType) -> np.ndarray

Normalize observations based on model type.

Parameters:

Name	Type	Description	Default
obs	ndarray	Raw observation array.	required
model_type	ModelType	Which normalization scheme to apply.	required

Returns:

Type	Description
ndarray	np.ndarray: Normalized observations.

predict

predict(cur_obs: ndarray, last_act: ndarray, obs_history: ndarray, act_history: ndarray) -> tuple[np.ndarray, np.ndarray]

Run inference through both networks.

Parameters:

Name	Type	Description	Default
cur_obs	ndarray	Current observation vector.	required
last_act	ndarray	Previous action vector.	required
obs_history	ndarray	Flattened observation history.	required
act_history	ndarray	Flattened action history.	required

Returns:

Type	Description
tuple[ndarray, ndarray]	tuple[np.ndarray, np.ndarray]: Normalized and raw actions.

Xadap_NN_control

Xadap_NN_control(params: VehicleParams = VehicleParams(), control_mask: list[int] = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], gains: dict | None = None)

Bases: FlightController

Controller combining high-level controller with neural adaptation.

Initialize the hybrid controller and buffers.

Parameters:

Name	Type	Description	Default
params	VehicleParams	Vehicle parameters.	VehicleParams()
control_mask	list[int]	Control mask for target/state channels.	[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
gains	dict | None	Optional PID gains passed to FlightController.	None

Methods:

Name	Description
update	Compute thrust and moments from target and state.
reset	Reset controller state and buffers.

Attributes:

Name	Type	Description
params	VehicleParams	Vehicle parameters for the controller.
control_mask	ndarray | Tensor	unified control mask for target/state channels, used in update(). e.g. using 9~11 for attitude yaw control.

params instance-attribute

params: VehicleParams = params

Vehicle parameters for the controller.

control_mask instance-attribute

control_mask: ndarray | Tensor = control_mask

unified control mask for target/state channels, used in update(). e.g. using 9~11 for attitude yaw control.

update

update(target: ndarray, state: ndarray) -> np.ndarray

Compute thrust and moments from target and state.

Parameters:

Name	Type	Description	Default
target	ndarray	Desired state vector.	required
state	ndarray	Current state vector.	required

Returns:

Type	Description
ndarray	np.ndarray: Control output ordered as [thrust, Mx, My, Mz].

reset

reset(*args, **kwargs)

Reset controller state and buffers.

dynamics

Dynamics models for LAV2.

Modules:

Name	Description
base	Base interfaces for LAV2 dynamics models.
params	Vehicle parameter definitions for LAV2.
rotor	Rotor dynamics model for thrust and torque outputs.
torch	Torch-based dynamics models for LAV2.
track	Track dynamics model for ground locomotion.

base

Base interfaces for LAV2 dynamics models.

Classes:

Name	Description
DynamicsBase	Abstract dynamics base class.

DynamicsBase

DynamicsBase(params: VehicleParams)

Bases: ABC

Abstract dynamics base class.

Initialize dynamics with vehicle parameters.

Methods:

Name	Description
update	Advance dynamics by one step and return outputs.
reset	Reset internal state to initial conditions.

Attributes:

Name	Type	Description
params	VehicleParams	Vehicle parameters for the dynamics model.
commands	ndarray | Tensor	Latest control commands applied to the dynamics model.

params instance-attribute

params: VehicleParams = params

Vehicle parameters for the dynamics model.

commands instance-attribute

commands: ndarray | Tensor

Latest control commands applied to the dynamics model.

update abstractmethod

update(*args, **kwargs)

Advance dynamics by one step and return outputs.

reset abstractmethod

reset(*args, **kwargs)

Reset internal state to initial conditions.

params

Vehicle parameter definitions for LAV2.

Classes:

Name	Description
VehicleParams	LAV2 vehicle parameters.

VehicleParams dataclass

VehicleParams(rho: float = 1.184, g: float = 9.81, sim_dt: float = 0.01, step_dt: float = 0.01, randomization: dict[str, dict[str, tuple[float, float]]] = dict(), mass: float = 2.1, inertia: list = (lambda: np.diag([0.11, 0.108, 0.215]).tolist())(), diameter: float = 0.476, Cdx: float = 0.5, x: float = 0.0952, y: float = 0.114423, h: float = -0.0125, alpha: float = 8 * (np.pi / 180), r_p: float = 0.0775, J_m: float = 1e-05, Ct: float = 0.666, Cq: float = 0.0716, theta0: float = 14.6 * (np.pi / 180), thetaTip: float = 6.8 * (np.pi / 180), lock: float = 0.6051, max_rpm: float = 10000.0, init_rpm: float = -1.0, rotor_rpm_rate_limit: float = 50000.0, tau_f: float = 0.01, tau_m: float = 0.05, tau_up: float = -1.0, tau_down: float = -1.0, r_w: float = 0.0285, n_w: int = 4, B: float = 0.136, L: float = 0.23, h_cg: float = 0.078, f: float = 0.01, mu0: float = 0.8, K: float = 20, track_cmd_rate_limit: float = 100, min_throttle_manual: float = 0.08, min_throttle: float = 0.12, max_throttle: float = 1.0, hover_throttle: float = -1)

LAV2 vehicle parameters.

Methods:

Name	Description
__post_init__	Calculate derived parameters.

Attributes:

Name	Type	Description
rho	float	空气密度 (kg/m^3)
g	float	重力加速度 (m/s^2)
sim_dt	float	仿真步长 (s)
step_dt	float	控制步长 (s)
randomization	dict[str, dict[str, tuple[float, float]]]	运行时参数随机化配置，按组件 (e.g. rotor/mixer) 划分为多个 scale 相对范围子表
mass	float	质量 (kg)
inertia	list	惯性矩阵 (kg*m^2)
diameter	float	直径 (m)
Cdx	float	空气阻力系数
x	float	x方向臂长 (m)
y	float	y方向臂长 (m)
h	float	重心高度 (m)
alpha	float	y方向涵道外倾角度 (rad)
r_p	float	螺旋桨半径 (m)
J_m	float	电机转动惯量 (kg*m^2)
Ct	float	推力系数
Cq	float	扭矩系数
theta0	float	桨根角 (rad)
thetaTip	float	桨尖角 (rad)
theta1	float	桨叶扭转角 (rad)
lock	float	Lock number
max_rpm	float	最大转速
init_rpm	float	初始转速
rotor_rpm_rate_limit	float	旋翼转速变化率限制 (rpm/s)
cT	float	转速平方到推力增益 (N/(rpm^2))
cM	float	转速平方到扭矩增益 (N*m/(rpm^2))
tau_f	float	电机滤波时间常数 (s)
tau_m	float	兼容保留的电机时间常数 (s)
tau_up	float	电机升速时间常数 (s)
tau_down	float	电机降速时间常数 (s)
r_w	float	主动轮半径 (m)
n_w	int	负重轮个数
B	float	两侧履带中心距 (m)
L	float	履带节点接地部分长度 (m)
h_cg	float	车体质心高度 (m)
f	float	滚动阻力系数(以干燥混凝土路面或沥青路面为例)
mu0	float	滑转率为1时的摩擦系数
K	float	依赖于土壤黏聚系数和摩擦特性的常数
track_cmd_rate_limit	float	履带主动轮指令变化率限制 (rad/s^2)
min_throttle_manual	float	MPC_MANTHR_MIN: Minimum manual thrust (0.0 - 1.0). Default: 0.08
min_throttle	float	MPC_THR_MIN: Minimum thrust in auto thrust control (0.05 - 1.0). Default: 0.12
max_throttle	float	MPC_THR_MAX: Maximum thrust in auto thrust control (0.0 - 1.0). Default: 1.0
hover_throttle	float	MPC_THR_HOVER: Hover throttle manual (0.1 - 0.8). Default: 0.5 in PX4, calculated hover throttle in LAV2

rho class-attribute instance-attribute

rho: float = 1.184

空气密度 (kg/m^3)

g class-attribute instance-attribute

g: float = 9.81

重力加速度 (m/s^2)

sim_dt class-attribute instance-attribute

sim_dt: float = 0.01

仿真步长 (s)

step_dt class-attribute instance-attribute

step_dt: float = 0.01

控制步长 (s)

randomization class-attribute instance-attribute

randomization: dict[str, dict[str, tuple[float, float]]] = field(default_factory=dict)

运行时参数随机化配置，按组件 (e.g. rotor/mixer) 划分为多个 scale 相对范围子表

mass class-attribute instance-attribute

mass: float = 2.1

质量 (kg)

inertia class-attribute instance-attribute

inertia: list = field(default_factory=lambda: tolist())

惯性矩阵 (kg*m^2)

diameter class-attribute instance-attribute

diameter: float = 0.476

直径 (m)

Cdx class-attribute instance-attribute

Cdx: float = 0.5

空气阻力系数

x class-attribute instance-attribute

x: float = 0.0952

x方向臂长 (m)

y class-attribute instance-attribute

y: float = 0.114423

y方向臂长 (m)

h class-attribute instance-attribute

h: float = -0.0125

重心高度 (m)

alpha class-attribute instance-attribute

alpha: float = 8 * (pi / 180)

y方向涵道外倾角度 (rad)

r_p class-attribute instance-attribute

r_p: float = 0.0775

螺旋桨半径 (m)

J_m class-attribute instance-attribute

J_m: float = 1e-05

电机转动惯量 (kg*m^2)

Ct class-attribute instance-attribute

Ct: float = 0.666

推力系数

Cq class-attribute instance-attribute

Cq: float = 0.0716

扭矩系数

theta0 class-attribute instance-attribute

theta0: float = 14.6 * (pi / 180)

桨根角 (rad)

thetaTip class-attribute instance-attribute

thetaTip: float = 6.8 * (pi / 180)

桨尖角 (rad)

theta1 class-attribute instance-attribute

theta1: float = field(init=False)

桨叶扭转角 (rad)

lock class-attribute instance-attribute

lock: float = 0.6051

Lock number

max_rpm class-attribute instance-attribute

max_rpm: float = 10000.0

最大转速

init_rpm class-attribute instance-attribute

init_rpm: float = -1.0

初始转速

rotor_rpm_rate_limit class-attribute instance-attribute

rotor_rpm_rate_limit: float = 50000.0

旋翼转速变化率限制 (rpm/s)

cT class-attribute instance-attribute

cT: float = field(init=False)

转速平方到推力增益 (N/(rpm^2))

cM class-attribute instance-attribute

cM: float = field(init=False)

转速平方到扭矩增益 (N*m/(rpm^2))

tau_f class-attribute instance-attribute

tau_f: float = 0.01

电机滤波时间常数 (s)

tau_m class-attribute instance-attribute

tau_m: float = 0.05

兼容保留的电机时间常数 (s)

tau_up class-attribute instance-attribute

tau_up: float = -1.0

电机升速时间常数 (s)

tau_down class-attribute instance-attribute

tau_down: float = -1.0

电机降速时间常数 (s)

r_w class-attribute instance-attribute

r_w: float = 0.0285

主动轮半径 (m)

n_w class-attribute instance-attribute

n_w: int = 4

负重轮个数

B class-attribute instance-attribute

B: float = 0.136

两侧履带中心距 (m)

L class-attribute instance-attribute

L: float = 0.23

履带节点接地部分长度 (m)

h_cg class-attribute instance-attribute

h_cg: float = 0.078

车体质心高度 (m)

f class-attribute instance-attribute

f: float = 0.01

滚动阻力系数(以干燥混凝土路面或沥青路面为例)

mu0 class-attribute instance-attribute

mu0: float = 0.8

滑转率为1时的摩擦系数

K class-attribute instance-attribute

K: float = 20

依赖于土壤黏聚系数和摩擦特性的常数

track_cmd_rate_limit class-attribute instance-attribute

track_cmd_rate_limit: float = 100

履带主动轮指令变化率限制 (rad/s^2)

min_throttle_manual class-attribute instance-attribute

min_throttle_manual: float = 0.08

MPC_MANTHR_MIN: Minimum manual thrust (0.0 - 1.0). Default: 0.08

min_throttle class-attribute instance-attribute

min_throttle: float = 0.12

MPC_THR_MIN: Minimum thrust in auto thrust control (0.05 - 1.0). Default: 0.12

max_throttle class-attribute instance-attribute

max_throttle: float = 1.0

MPC_THR_MAX: Maximum thrust in auto thrust control (0.0 - 1.0). Default: 1.0

hover_throttle class-attribute instance-attribute

hover_throttle: float = -1

MPC_THR_HOVER: Hover throttle manual (0.1 - 0.8). Default: 0.5 in PX4, calculated hover throttle in LAV2

__post_init__

__post_init__()

Calculate derived parameters.

rotor

Rotor dynamics model for thrust and torque outputs.

Classes:

Name	Description
RotorDynamics	Motor dynamics model for rotor thrust and torque.

RotorDynamics

RotorDynamics(params: VehicleParams = VehicleParams())

Bases: DynamicsBase

Motor dynamics model for rotor thrust and torque.

References

• Aerial Gym: https://github.com/ntnu-arl/aerial_gym_simulator/blob/main/aerial_gym/control/motor_model.py
• MARL_cooperative_aerial_manipulation_ext: https://github.com/Aerial-Manipulation-Lab/MARL_cooperative_aerial_manipulation_ext/blob/main/exts/MARL_mav_carry_ext/MARL_mav_carry_ext/controllers/motor_model.py
• isaac_drone_racer: https://github.com/kousheekc/isaac_drone_racer/blob/master/dynamics/motor.py

Initialize rotor dynamics and rate-limit parameters.

Methods:

Name	Description
update	Calculate the output thrust and torque for each motor based on the motor commands.
reset	Reset the rotor model to initial conditions.

Attributes:

Name	Type	Description
params	VehicleParams	Vehicle parameters for the dynamics model.

params instance-attribute

params: VehicleParams = params

Vehicle parameters for the dynamics model.

update

update(commands: ndarray) -> np.ndarray

Calculate the output thrust and torque for each motor based on the motor commands.

Parameters:

Name	Type	Description	Default
commands	ndarray	Command array for each motor. Shape: (4,)	required

Returns:

Type	Description
ndarray	np.ndarray: Thrust and torque array for each motor. Shape: (8,)

reset

reset()

Reset the rotor model to initial conditions.

torch

Torch-based dynamics models for LAV2.

Modules:

Name	Description
rotor	Torch rotor dynamics for batched simulation.
track	Torch track dynamics for batched simulation.

rotor

Torch rotor dynamics for batched simulation.

Classes:

Name	Description
RotorDynamics	Batched rotor dynamics model.

RotorDynamics

RotorDynamics(params: VehicleParams = VehicleParams(), num_envs: int = 1, device: str | device = 'cpu')

Bases: DynamicsBase

Batched rotor dynamics model.

Initialize batched rotor dynamics on the specified device.

Methods:

Name	Description
randomize	Randomize rotor runtime parameters from nominal values.
update	Calculate the output thrust and torque for each motor based on the motor commands.
reset	Resets the rotor model to initial conditions.

Attributes:

Name	Type	Description
params	VehicleParams	Vehicle parameters for the dynamics model.

params instance-attribute

params: VehicleParams = params

Vehicle parameters for the dynamics model.

randomize

randomize(env_ids: slice | Sequence[int] | Tensor | None = None, randomization: dict[str, tuple[float, float]] | None = None) -> None

Randomize rotor runtime parameters from nominal values.

update

update(commands: Tensor) -> torch.Tensor

Calculate the output thrust and torque for each motor based on the motor commands.

Parameters:

Name	Type	Description	Default
commands	Tensor	Command tensor for each motor. Shape: (num_envs, 4)	required

Returns:

Type	Description
Tensor	torch.Tensor: Thrust and torque tensor for each motor. Shape: (num_envs, 8)

reset

reset(env_ids: slice | Sequence[int] | Tensor) -> None

Resets the rotor model to initial conditions.

Parameters:

Name	Type	Description	Default
env_ids	slice | Sequence[int] | Tensor	The environment ids to reset.	required

track

Torch track dynamics for batched simulation.

Classes:

Name	Description
TrackDynamics	Batched track dynamics model.

TrackDynamics

TrackDynamics(params: VehicleParams = VehicleParams(), num_envs: int = 1, device: str | device = 'cpu')

Bases: DynamicsBase

Batched track dynamics model.

Initialize batched track dynamics on the specified device.

Methods:

Name	Description
update	Compute track forces for batched environments.
reset	Resets the track model to initial conditions.

Attributes:

Name	Type	Description
params	VehicleParams	Vehicle parameters for the dynamics model.

params instance-attribute

params: VehicleParams = params

Vehicle parameters for the dynamics model.

update

update(commands: Tensor, state: Tensor) -> torch.Tensor

Compute track forces for batched environments.

Parameters:

Name	Type	Description	Default
commands	Tensor	Shape (num_envs, 2) track commands [left, right].	required
state	Tensor	Shape (num_envs, 3) body-frame velocities [u, v, r].	required

Returns:

Type	Description
Tensor	torch.Tensor: Tensor of shape (num_envs, 2, n_w, 2) [side, wheel, (Fx, Fy)].

reset

reset(env_ids: slice | Sequence[int] | Tensor) -> None

Resets the track model to initial conditions.

Parameters:

Name	Type	Description	Default
env_ids	slice | Sequence[int] | Tensor	The environment ids to reset.	required

track

Track dynamics model for ground locomotion.

Classes:

Name	Description
TrackDynamics	Track dynamics model for left/right traction forces.

TrackDynamics

TrackDynamics(params: VehicleParams = VehicleParams())

Bases: DynamicsBase

Track dynamics model for left/right traction forces.

Initialize track dynamics geometry and normal load distribution.

Methods:

Name	Description
update	Calculate the output force of the tracks based on track commands.
reset	Reset the track model to initial conditions.

Attributes:

Name	Type	Description
params	VehicleParams	Vehicle parameters for the dynamics model.

params instance-attribute

params: VehicleParams = params

Vehicle parameters for the dynamics model.

update

update(commands: ndarray, state: ndarray) -> np.ndarray

Calculate the output force of the tracks based on track commands.

Parameters:

Name	Type	Description	Default
commands	ndarray	Command array for each track. Shape: (2,)	required
state	ndarray	Current vehicle state array. Shape: (3,)	required

Returns:

Type	Description
ndarray	np.ndarray: Force array for each track. Shape: (2, n_w, 2) [side, wheel, (Fx, Fy)]

reset

reset()

Reset the track model to initial conditions.

gamepads

Gamepad input helpers for LAV2.

Modules:

Name	Description
common	Shared data structures for gamepad inputs.
sdl2	SDL2 gamepad backend for LAV2.

Classes:

Name	Description
Key	Gamepad input abstraction shared across backends.
ControllerEventLoop	Minimal SDL2 controller event loop for LAV2.

Functions:

Name	Description
controller_key_from_event	Convert an SDL2 event to a Key.

Key dataclass

Key(keytype: str, index: int, name: str | None = None, value: float | None = None)

Gamepad input abstraction shared across backends.

ControllerEventLoop

ControllerEventLoop(handle_key: Callable[[Key], None], alive: Event | None = None, timeout: int = 2000)

Minimal SDL2 controller event loop for LAV2.

Initialize the controller event loop.

Parameters:

Name	Type	Description	Default
handle_key	Callable[[Key], None]	Callback for parsed keys.	required
alive	Event | None	Control event loop lifetime.	None
timeout	int	SDL wait timeout in milliseconds.	2000

Methods:

Name	Description
stop	Stop the event loop.
run	Run the SDL2 event loop until stopped.

stop

stop() -> None

Stop the event loop.

run

run() -> None

Run the SDL2 event loop until stopped.

controller_key_from_event

controller_key_from_event(event: SDL_Event, deadzone: float = 0.05) -> Key | None

Convert an SDL2 event to a Key.

Parameters:

Name	Type	Description	Default
event	SDL_Event	Raw SDL2 event.	required
deadzone	float	Deadzone for analog axes.	0.05

Returns:

Type	Description
Key | None	Key | None: Parsed key input, if any.

common

Shared data structures for gamepad inputs.

Classes:

Name	Description
Key	Gamepad input abstraction shared across backends.

Key dataclass

Key(keytype: str, index: int, name: str | None = None, value: float | None = None)

Gamepad input abstraction shared across backends.

sdl2

SDL2 gamepad backend for LAV2.

Classes:

Name	Description
ControllerEventLoop	Minimal SDL2 controller event loop for LAV2.

Functions:

Name	Description
controller_key_from_event	Convert an SDL2 event to a Key.

ControllerEventLoop

ControllerEventLoop(handle_key: Callable[[Key], None], alive: Event | None = None, timeout: int = 2000)

Minimal SDL2 controller event loop for LAV2.

Initialize the controller event loop.

Parameters:

Name	Type	Description	Default
handle_key	Callable[[Key], None]	Callback for parsed keys.	required
alive	Event | None	Control event loop lifetime.	None
timeout	int	SDL wait timeout in milliseconds.	2000

Methods:

Name	Description
stop	Stop the event loop.
run	Run the SDL2 event loop until stopped.

stop

stop() -> None

Stop the event loop.

run

run() -> None

Run the SDL2 event loop until stopped.

controller_key_from_event

controller_key_from_event(event: SDL_Event, deadzone: float = 0.05) -> Key | None

Convert an SDL2 event to a Key.

Parameters:

Name	Type	Description	Default
event	SDL_Event	Raw SDL2 event.	required
deadzone	float	Deadzone for analog axes.	0.05

Returns:

Type	Description
Key | None	Key | None: Parsed key input, if any.

runner

Runner utilities for LAV2 training and evaluation.

Modules:

Name	Description
common	Shared utilities for common runner helpers and play scripts.
skrl	SKRL runner helpers for LAV2.

common

Shared utilities for common runner helpers and play scripts.

Modules:

Name	Description
mujoco	Modular play utilities for LAV2 policy deployment.

mujoco

Modular play utilities for LAV2 policy deployment.

This module provides reusable components for playing trained policies in MuJoCo environments with keyboard and gamepad control, supporting various action and command modes.

Design follows lav2.controller.run pattern.

Classes:

Name	Description
MuJoCoPlayConfig	Configuration for MuJoCo policy play session.
MuJoCoPlayRunner	Modular runner for MuJoCo policy play sessions with keyboard/gamepad control.

MuJoCoPlayConfig dataclass

MuJoCoPlayConfig(checkpoint_path: str, params: VehicleParams = VehicleParams(), decimation: int = 2, action_mode: str = 'cmd_motor_speeds', command_mode: str = 'tgt_pos', initial_target: ndarray = (lambda: np.array([1.0, 1.0, 1.5, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]))(), mjcf_path: str = str(files('lav2.assets').joinpath('mjcf/scene.xml')), enable_gamepad: bool = False, trajectory: BaseTrajectory | None = None, pos_inc: tuple[float, float, float] = (1.0, 1.0, 0.5), vel_limits: tuple[float, float, float] = (5.0, 5.0, 5.0), att_limits: tuple[float, float, float] = (1.0, 1.0, np.pi), br_limits: tuple[float, float, float] = (2.0, 2.0, 2.0))

Configuration for MuJoCo policy play session.

Attributes:

Name	Type	Description
checkpoint_path	str	Path to the policy checkpoint (JIT or ONNX).
params	VehicleParams	Vehicle parameters (auto-created if None).
decimation	int	Control decimation factor.
action_mode	str	Action mode - 'cmd_pos', 'cmd_vel', 'cmd_ctatt', 'cmd_ctbr', 'cmd_ctbm', 'cmd_motor_speeds'.
command_mode	str	Command mode - 'tgt_pos', 'tgt_vel', 'tgt_ctatt', 'tgt_ctbr'.
initial_target	ndarray	Initial target [x,y,z,vx,vy,vz,roll,pitch,yaw,wx,wy,wz].
mjcf_path	str	Path to MuJoCo XML scene file.
enable_gamepad	bool	Enable gamepad input.
pos_inc	tuple[float, float, float]	Positional increment for keyboard control (dx, dy, dz).
vel_limits	tuple[float, float, float]	Velocity limits [vx, vy, vz].
att_limits	tuple[float, float, float]	Attitude limits [roll, pitch, yaw].
br_limits	tuple[float, float, float]	Body rate limits [wx, wy, wz].

Methods:

Name	Description
__post_init__	Validate configuration values after initialization.

__post_init__

__post_init__()

Validate configuration values after initialization.

Raises:

Type	Description
ValueError	If action_mode or command_mode is not a valid option.

MuJoCoPlayRunner

MuJoCoPlayRunner(cfg: MuJoCoPlayConfig, policy_runner: JITPolicyRunner | ONNXPolicyRunner, obs_extractor: Callable[[MjData, ndarray, ndarray], ndarray] | None = None)

Modular runner for MuJoCo policy play sessions with keyboard/gamepad control.

Provides reusable observation extraction, control loop, and input handling for LAV2 flight control in MuJoCo. Supports multiple action modes following the lav2.controller.run design pattern.

Initialize the play runner.

Parameters:

Name	Type	Description	Default
cfg	MuJoCoPlayConfig	Play cfguration.	required
policy_runner	JITPolicyRunner | ONNXPolicyRunner	Policy runner (JIT or ONNX).	required
obs_extractor	Callable[[MjData, ndarray, ndarray], ndarray] | None	Custom observation extraction function. If None, uses default flight observation extractor. Signature: (d: MjData, target: ndarray, last_action: ndarray) -> obs	None

Methods:

Name	Description
key_callback	Handle keyboard input to update target position (tgt_pos mode only).
gamepad_callback	Handle gamepad input following lav2.controller.run pattern.
process_actions	Process policy action and compute RPM commands (control loop).
apply_actions	Apply rotor dynamics and update MuJoCo actuators (simulation loop).
control_callback	Control loop callback for MuJoCo viewer (respects decimation).
load_callback	Load MuJoCo model and data.
run	Launch the MuJoCo viewer and start the play session.

key_callback

key_callback(keycode: int)

Handle keyboard input to update target position (tgt_pos mode only).

Keys 1-8 adjust target position and heading. Note: Currently only supports tgt_pos command mode.

gamepad_callback

gamepad_callback(key: Key) -> None

Handle gamepad input following lav2.controller.run pattern.

Parameters:

Name	Type	Description	Default
key	Key	Gamepad key event (axis or button).	required

process_actions

process_actions(m: MjModel, d: MjData)

Process policy action and compute RPM commands (control loop).

This method is called at the decimated control rate.

Parameters:

Name	Type	Description	Default
m	MjModel	MuJoCo model.	required
d	MjData	MuJoCo data.	required

apply_actions

apply_actions(m: MjModel, d: MjData)

Apply rotor dynamics and update MuJoCo actuators (simulation loop).

This runs at every simulation step, not subject to decimation.

Parameters:

Name	Type	Description	Default
m	MjModel	MuJoCo model.	required
d	MjData	MuJoCo data.	required

control_callback

control_callback(m: MjModel, d: MjData)

Control loop callback for MuJoCo viewer (respects decimation).

Executes policy inference and command processing, but applies rotor dynamics separately to respect simulation decimation.

Parameters:

Name	Type	Description	Default
m	MjModel	MuJoCo model.	required
d	MjData	MuJoCo data.	required

load_callback

load_callback() -> tuple[mujoco.MjModel, mujoco.MjData]

Load MuJoCo model and data.

Returns:

Type	Description
tuple[MjModel, MjData]	tuple[mujoco.MjModel, mujoco.MjData]: Tuple of (model, data).

run

run()

Launch the MuJoCo viewer and start the play session.

skrl

SKRL runner helpers for LAV2.

Modules:

Name	Description
__main__	CLI entry point for SKRL configuration runners.
cfg	SKRL configuration modules for LAV2.
eval	Evaluation helpers for SKRL policies.
isaaclab	SKRL runner for Isaac Lab tasks.

__main__

CLI entry point for SKRL configuration runners.

Functions:

Name	Description
get_available_configs	Return available configuration module names.
main	Run a selected SKRL configuration script.

get_available_configs

get_available_configs()

Return available configuration module names.

main

main()

Run a selected SKRL configuration script.

cfg

SKRL configuration modules for LAV2.

Modules:

Name	Description
LAV2_base	SKRL configuration for the LAV2 base task.
LAV2_base_moe	SKRL configuration for the LAV2 base MoE task.
LAV2_base_vel	SKRL configuration for the LAV2 base velocity task.
LAV2_navrl	SKRL configuration for the LAV2 NavRL task.

LAV2_base

SKRL configuration for the LAV2 base task.

Classes:

Name	Description
Shared	Shared actor-critic model for PPO.

Functions:

Name	Description
get_agent_cfg	Return the PPO agent configuration.
get_model_class	Return the model class for the agent.
get_agent_class	Return the agent class.
get_memory_class	Return the memory class.
get_memory_cfg	Return the memory configuration.
get_trainer_class	Return the trainer class.
get_trainer_cfg	Return the trainer configuration.

Shared

Shared(observation_space, state_space, action_space, device, clip_actions=False, clip_log_std=True, min_log_std=-20, max_log_std=2, reduction='sum')

Bases: GaussianMixin, DeterministicMixin, Model

Shared actor-critic model for PPO.

Initialize shared model layers and mixins.

Methods:

Name	Description
act	Compute actions or values depending on role.
compute	Compute policy/value outputs depending on role.

act

act(inputs: Any, role: str) -> Any

Compute actions or values depending on role.

Returns:

Type	Description
Any	tuple | None: Action outputs for the given role, or None if role is unknown.

compute

compute(inputs: Any, role: str) -> Any

Compute policy/value outputs depending on role.

Returns:

Type	Description
Any	tuple | None: (output_tensor, extras_dict) for policy/value roles, or None.

get_agent_cfg

get_agent_cfg() -> Any

Return the PPO agent configuration.

get_model_class

get_model_class() -> Any

Return the model class for the agent.

get_agent_class

get_agent_class() -> Any

Return the agent class.

get_memory_class

get_memory_class() -> Any

Return the memory class.

get_memory_cfg

get_memory_cfg() -> Any

Return the memory configuration.

get_trainer_class

get_trainer_class() -> Any

Return the trainer class.

get_trainer_cfg

get_trainer_cfg() -> Any

Return the trainer configuration.

LAV2_base_moe

SKRL configuration for the LAV2 base MoE task.

Classes:

Name	Description
PPO_MoE	PPO variant with MoE bias updates.
MoELayer	Mixture-of-experts layer with top-k routing.
Shared	Shared actor-critic model with MoE trunk.

Functions:

Name	Description
get_agent_cfg	Return the PPO agent configuration.
get_model_class	Return the model class for the agent.
get_agent_class	Return the agent class.
get_memory_class	Return the memory class.
get_memory_cfg	Return the memory configuration.
get_trainer_class	Return the trainer class.
get_trainer_cfg	Return the trainer configuration.

PPO_MoE

Bases: PPO

PPO variant with MoE bias updates.

Methods:

Name	Description
update	Run PPO update and then update MoE biases.

update

update(*args: Any, **kwargs: Any) -> None

Run PPO update and then update MoE biases.

MoELayer

MoELayer(input_size, output_size, num_experts, k)

Bases: Module

Mixture-of-experts layer with top-k routing.

Initialize experts, gate network, and load-balancing bias.

Methods:

Name	Description
forward	Compute routed expert outputs.

forward

forward(x: Tensor) -> torch.Tensor

Compute routed expert outputs.

Returns:

Type	Description
Tensor	torch.Tensor: Weighted combination of top-k expert outputs.

Shared

Shared(observation_space, state_space, action_space, device, clip_actions=False, clip_log_std=True, min_log_std=-20, max_log_std=2, reduction='sum')

Bases: GaussianMixin, DeterministicMixin, Model

Shared actor-critic model with MoE trunk.

Initialize the shared MoE actor-critic model.

Methods:

Name	Description
act	Compute actions or values depending on role.
compute	Compute policy/value outputs depending on role.

act

act(inputs: Any, role: str) -> Any

Compute actions or values depending on role.

Returns:

Type	Description
Any	tuple | None: Action outputs for the given role, or None if role is unknown.

compute

compute(inputs: Any, role: str) -> Any

Compute policy/value outputs depending on role.

Returns:

Type	Description
Any	tuple | None: (output_tensor, extras_dict) for policy/value roles, or None.

get_agent_cfg

get_agent_cfg() -> Any

Return the PPO agent configuration.

get_model_class

get_model_class() -> Any

Return the model class for the agent.

get_agent_class

get_agent_class() -> Any

Return the agent class.

get_memory_class

get_memory_class() -> Any

Return the memory class.

get_memory_cfg

get_memory_cfg() -> Any

Return the memory configuration.

get_trainer_class

get_trainer_class() -> Any

Return the trainer class.

get_trainer_cfg

get_trainer_cfg() -> Any

Return the trainer configuration.

LAV2_base_vel

SKRL configuration for the LAV2 base velocity task.

Classes:

Name	Description
Shared	Shared actor-critic model for PPO.

Functions:

Name	Description
get_agent_cfg	Return the PPO agent configuration.
get_model_class	Return the model class for the agent.
get_agent_class	Return the agent class.
get_memory_class	Return the memory class.
get_memory_cfg	Return the memory configuration.
get_trainer_class	Return the trainer class.
get_trainer_cfg	Return the trainer configuration.

Shared

Shared(observation_space, state_space, action_space, device, clip_actions=False, clip_log_std=True, min_log_std=-20, max_log_std=2, reduction='sum')

Bases: GaussianMixin, DeterministicMixin, Model

Shared actor-critic model for PPO.

Initialize shared model layers and mixins.

Methods:

Name	Description
act	Compute actions or values depending on role.
compute	Compute policy/value outputs depending on role.

act

act(inputs: Any, role: str) -> Any

Compute actions or values depending on role.

Returns:

Type	Description
Any	tuple | None: Action outputs for the given role, or None if role is unknown.

compute

compute(inputs: Any, role: str) -> Any

Compute policy/value outputs depending on role.

Returns:

Type	Description
Any	tuple | None: (output_tensor, extras_dict) for policy/value roles, or None.

get_agent_cfg

get_agent_cfg() -> Any

Return the PPO agent configuration.

get_model_class

get_model_class() -> Any

Return the model class for the agent.

get_agent_class

get_agent_class() -> Any

Return the agent class.

get_memory_class

get_memory_class() -> Any

Return the memory class.

get_memory_cfg

get_memory_cfg() -> Any

Return the memory configuration.

get_trainer_class

get_trainer_class() -> Any

Return the trainer class.

get_trainer_cfg

get_trainer_cfg() -> Any

Return the trainer configuration.

LAV2_navrl

SKRL configuration for the LAV2 NavRL task.

Classes:

Name	Description
Shared	Shared actor-critic model with sensor fusion.

Functions:

Name	Description
get_agent_cfg	Return the PPO agent configuration.
get_model_class	Return the model class for the agent.
get_agent_class	Return the agent class.
get_memory_class	Return the memory class.
get_memory_cfg	Return the memory configuration.
get_trainer_class	Return the trainer class.
get_trainer_cfg	Return the trainer configuration.

Shared

Shared(observation_space, state_space, action_space, device, clip_actions=False, clip_log_std=True, min_log_std=-20, max_log_std=2, reduction='sum')

Bases: GaussianMixin, DeterministicMixin, Model

Shared actor-critic model with sensor fusion.

Initialize encoders and policy/value heads.

Methods:

Name	Description
act	Compute actions or values depending on role.
compute	Compute policy/value outputs depending on role.

act

act(inputs: Any, role: str) -> Any

Compute actions or values depending on role.

Returns:

Type	Description
Any	tuple | None: Action outputs for the given role, or None if role is unknown.

compute

compute(inputs: Any, role: str) -> Any

Compute policy/value outputs depending on role.

Returns:

Type	Description
Any	tuple | None: (output_tensor, extras_dict) for policy/value roles, or None.

get_agent_cfg

get_agent_cfg() -> Any

Return the PPO agent configuration.

get_model_class

get_model_class() -> Any

Return the model class for the agent.

get_agent_class

get_agent_class() -> Any

Return the agent class.

get_memory_class

get_memory_class() -> Any

Return the memory class.

get_memory_cfg

get_memory_cfg() -> Any

Return the memory configuration.

get_trainer_class

get_trainer_class() -> Any

Return the trainer class.

get_trainer_cfg

get_trainer_cfg() -> Any

Return the trainer configuration.

eval

Evaluation helpers for SKRL policies.

Classes:

Name	Description
JITPolicyRunner	TorchScript policy runner.
ONNXPolicyRunner	ONNX policy runner.

Functions:

Name	Description
load_agent	Load a trained agent for evaluation/inference.
load_policy	Load a policy as a unified runner supporting JIT or ONNX.

JITPolicyRunner

JITPolicyRunner(model_path: str, device: str | device)

Bases: _BasePolicyRunner

TorchScript policy runner.

Initialize a TorchScript policy runner.

Parameters:

Name	Type	Description	Default
model_path	str	Path to the JIT model.	required
device	str | device	Target device.	required

Methods:

Name	Description
act	Compute actions from observations.

act

act(observations: ndarray | Tensor) -> np.ndarray

Compute actions from observations.

Returns:

Type	Description
ndarray	np.ndarray: Action array with shape matching the model output.

ONNXPolicyRunner

ONNXPolicyRunner(model_path: str, providers: list | None = None)

Bases: _BasePolicyRunner

ONNX policy runner.

Initialize an ONNX policy runner.

Parameters:

Name	Type	Description	Default
model_path	str	Path to the ONNX model.	required
providers	list | None	Optional ONNX providers list.	None

Raises:

Type	Description
RuntimeError	If the ONNX model has no inputs.

Methods:

Name	Description
act	Compute actions from observations.

act

act(observations: ndarray | Tensor) -> np.ndarray

Compute actions from observations.

Returns:

Type	Description
ndarray	np.ndarray: Action array with shape matching the model output.

load_agent

load_agent(model_cls: type, agent_cfg: Any, checkpoint_path: str, observation_shape: tuple, action_shape: tuple, state_shape: tuple | None = None, device: str | device | None = None) -> tuple

Load a trained agent for evaluation/inference.

Parameters:

Name	Type	Description	Default
model_cls	type	Class of the model (policy).	required
agent_cfg	Any	Agent configuration object.	required
checkpoint_path	str	Path to the checkpoint file.	required
observation_shape	tuple	Tuple defining observation dimensions.	required
action_shape	tuple	Tuple defining action dimensions.	required
state_shape	tuple | None	Tuple defining state dimensions.	None
device	str | device | None	Torch device (auto-detected if None).	None

Returns:

Name	Type	Description
tuple	tuple	(policy, observation_preprocessor, device).

Raises:

Type	Description
FileNotFoundError	If the checkpoint file is not found.
ImportError	If gymnasium is not installed.

load_policy

load_policy(model_path: str, device: str | device = 'cpu') -> JITPolicyRunner | ONNXPolicyRunner

Load a policy as a unified runner supporting JIT or ONNX.

The returned object exposes act(observations) and returns numpy actions. Observations may be numpy arrays or torch tensors. A 1-D observation is treated as a single batch and the returned actions are 1-D accordingly.

Returns:

Type	Description
JITPolicyRunner | ONNXPolicyRunner	JITPolicyRunner | ONNXPolicyRunner: Loaded policy runner.

Raises:

Type	Description
ValueError	If model_path is None.

isaaclab

SKRL runner for Isaac Lab tasks.

Functions:

Name	Description
parse_args	Build the CLI argument parser.
run	Run training or evaluation for an Isaac Lab task.

parse_args

parse_args() -> argparse.ArgumentParser

Build the CLI argument parser.

Returns:

Type	Description
ArgumentParser	argparse.ArgumentParser: Configured argument parser.

run

run(task_name: str, model_cls: type, agent_cls: type, agent_cfg: Any, memory_cls: type, memory_cfg: dict, trainer_cls: type, trainer_cfg: dict, export_jit: bool = False, export_onnx: bool = False)

Run training or evaluation for an Isaac Lab task.

Parameters:

Name	Type	Description	Default
task_name	str	Isaac Lab task name.	required
model_cls	type	Policy model class.	required
agent_cls	type	Agent class.	required
agent_cfg	Any	Agent configuration object.	required
memory_cls	type	Memory class.	required
memory_cfg	dict	Memory configuration.	required
trainer_cls	type	Trainer class.	required
trainer_cfg	dict	Trainer configuration.	required
export_jit	bool	Whether to export JIT policy.	False
export_onnx	bool	Whether to export ONNX policy.	False

tasks

Task packages for supported training frameworks for LAV2.

Modules:

Name	Description
genesis_forge
isaaclab
mjlab

genesis_forge

Modules:

Name	Description
LAV2_base
LAV2_base_velocity

LAV2_base

Modules:

Name	Description
environment	LAV2 Base Environment using managers to handle everything.
eval
mdp

environment

LAV2 Base Environment using managers to handle everything.

Classes:

Name	Description
LAV2BaseEnv	Example training environment for the LAV2 robot.

LAV2BaseEnv

LAV2BaseEnv(num_envs: int = 1, dt: float = 1 / 100, max_episode_length_s: int | None = 10, headless: bool = True)

Bases: ManagedEnvironment

Example training environment for the LAV2 robot.

Methods:

Name	Description
config	Configure the environment managers

config

config()

Configure the environment managers

eval

Functions:

Name	Description
get_latest_model	Get the last model from the log directory

get_latest_model

get_latest_model(log_dir: str) -> str

Get the last model from the log directory

mdp

Modules:

Name	Description
actions
commands
rewards

Classes:

Name	Description
FlightActionMapper	Map batched normalized actions into controller targets, wrench, or RPM.
FlightMappingLimits	Per-channel limits used when mapping normalized actions to targets.
Mixer	Map thrust and moment demands to rotor commands.
FlightController	PID flight controller (batched).
VehicleParams	LAV2 vehicle parameters.
RotorDynamics	Batched rotor dynamics model.
ControlActionManager	Base class for action managers that control actuators.
PositionDebugVisualizerConfig	Defines the configuration for the debug visualizer.
PositionCommandManager	Generates a position command from uniform distribution.

Functions:

Name	Description
lin_vel_l2	Penalize base linear velocity using L2 squared kernel.
ang_vel_l2	Penalize base angular velocity using L2 squared kernel.
pos_error_l2	Penalize asset pos from its target pos using L2 squared kernel.
pos_error_tanh	Penalize asset pos from its target pos using tanh kernel.

FlightActionMapper

FlightActionMapper(control_mode: str, params: VehicleParams = VehicleParams(), limits: FlightMappingLimits = FlightMappingLimits(), *, num_envs: int = 1, device: str | device = 'cpu')

Map batched normalized actions into controller targets, wrench, or RPM.

Initialize a batched Torch flight action mapper for the selected mode.

Methods:

Name	Description
map_action	Map a normalized batched action into RPM commands and intermediates.

Attributes:

Name	Type	Description
requires_controller	bool	Whether this mode needs a controller target expansion step.

requires_controller property

requires_controller: bool

Whether this mode needs a controller target expansion step.

map_action

map_action(action: Tensor, mixer: Mixer, *, state: Tensor | None = None, flight_controller: ControllerBase | None = None) -> FlightMappingResult

Map a normalized batched action into RPM commands and intermediates.

FlightMappingLimits dataclass

FlightMappingLimits(vel_limits: tuple[float, float, float] = (5.0, 5.0, 5.0), att_limits: tuple[float, float, float] = (1.0, 1.0, math.pi), br_limits: tuple[float, float, float] = (2.0, 2.0, 2.0))

Per-channel limits used when mapping normalized actions to targets.

Mixer

Mixer(params: VehicleParams = VehicleParams(), num_envs: int = 1, device: str | device = 'cpu')

Map thrust and moment demands to rotor commands.

Note

电机推力分配： 电机1 = 推力 - 横滚(roll) + 俯仰(pitch) + 偏航(yaw) 电机2 = 推力 + 横滚(roll) - 俯仰(pitch) + 偏航(yaw) 电机3 = 推力 + 横滚(roll) + 俯仰(pitch) - 偏航(yaw) 电机4 = 推力 - 横滚(roll) - 俯仰(pitch) - 偏航(yaw)

Initialize the torch mixer with vehicle parameters and device.

Methods:

Name	Description
randomize	Randomize mixer runtime parameters from nominal values.
apply_thrust_curve	Apply thrust curve to normalized thrust command.
calculate_rotor_commands	Allocate thrust and moments into rotor speeds.

randomize

randomize(env_ids=None, randomization=None)

Randomize mixer runtime parameters from nominal values.

apply_thrust_curve

apply_thrust_curve(normalized_thrust: Tensor, mode: int = 1) -> torch.Tensor

Apply thrust curve to normalized thrust command.

References

cmd_ctatt: https://github.com/PX4/PX4-Autopilot/blob/e5a483e4b71ab4b994cb0121fe28992d8ce0c8cd/src/modules/mc_att_control/mc_att_control_main.cpp#L111-L137 cmd_ctbr: https://github.com/PX4/PX4-Autopilot/blob/e5a483e4b71ab4b994cb0121fe28992d8ce0c8cd/src/modules/mc_rate_control/MulticopterRateControl.cpp#L156-L179

Parameters:

Name	Type	Description	Default
normalized_thrust	Tensor	Normalized thrust command [-1, 1]. Shape: arbitrary, e.g. (num_envs, n_motors). Applied elementwise.	required
mode	int	Thrust curve mode.	1

Returns:

Type	Description
Tensor	torch.Tensor: Thrust in Newtons with the same shape as normalized_thrust.

Raises:

Type	Description
NotImplementedError	If the specified mode is not implemented.

calculate_rotor_commands

calculate_rotor_commands(control: Tensor) -> torch.Tensor

Allocate thrust and moments into rotor speeds.

Parameters:

Name	Type	Description	Default
control	Tensor	Desired total thrust and roll/pitch/yaw moments. Shape: (num_envs, 4)	required

Returns:

Type	Description
Tensor	torch.Tensor: The amplitude of each motor's rotational speed (calculated using sqrt(clip(rpm_sq))). Shape: (num_envs, 4)

FlightController

FlightController(params: VehicleParams = VehicleParams(), control_mask: list[int] = [1, 1, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0], gains: dict | None = None, limits: dict | None = None, num_envs: int = 1, device: str | device = 'cpu')

Bases: ControllerBase

PID flight controller (batched).

Initialize the batched flight PID controller.

Methods:

Name	Description
randomize	Randomize configured PID gains and command limits for selected environments.
reset	Reset PID loops for selected environments.
update	Update controller.

Attributes:

Name	Type	Description
params	VehicleParams	Vehicle parameters for the controller.
control_mask	ndarray | Tensor	unified control mask for target/state channels, used in update(). e.g. using 9~11 for attitude yaw control.

params instance-attribute

params: VehicleParams = params

Vehicle parameters for the controller.

control_mask instance-attribute

control_mask: ndarray | Tensor = control_mask

unified control mask for target/state channels, used in update(). e.g. using 9~11 for attitude yaw control.

randomize

randomize(env_ids=None, gains=None, limits=None)

Randomize configured PID gains and command limits for selected environments.

reset

reset(env_ids)

Reset PID loops for selected environments.

update

update(target: Tensor, state: Tensor) -> torch.Tensor

Update controller.

Parameters:

Name	Type	Description	Default
target	Tensor	(N, 12) desired [x, y, z, vx, vy, vz, roll, pitch, yaw, wx, wy, wz]	required
state	Tensor	(N, 12) current state same ordering.	required

Returns:

Type	Description
Tensor	torch.Tensor: Tensor of shape (N, 4): [thrust, roll_moment, pitch_moment, yaw_moment].

VehicleParams dataclass

VehicleParams(rho: float = 1.184, g: float = 9.81, sim_dt: float = 0.01, step_dt: float = 0.01, randomization: dict[str, dict[str, tuple[float, float]]] = dict(), mass: float = 2.1, inertia: list = (lambda: np.diag([0.11, 0.108, 0.215]).tolist())(), diameter: float = 0.476, Cdx: float = 0.5, x: float = 0.0952, y: float = 0.114423, h: float = -0.0125, alpha: float = 8 * (np.pi / 180), r_p: float = 0.0775, J_m: float = 1e-05, Ct: float = 0.666, Cq: float = 0.0716, theta0: float = 14.6 * (np.pi / 180), thetaTip: float = 6.8 * (np.pi / 180), lock: float = 0.6051, max_rpm: float = 10000.0, init_rpm: float = -1.0, rotor_rpm_rate_limit: float = 50000.0, tau_f: float = 0.01, tau_m: float = 0.05, tau_up: float = -1.0, tau_down: float = -1.0, r_w: float = 0.0285, n_w: int = 4, B: float = 0.136, L: float = 0.23, h_cg: float = 0.078, f: float = 0.01, mu0: float = 0.8, K: float = 20, track_cmd_rate_limit: float = 100, min_throttle_manual: float = 0.08, min_throttle: float = 0.12, max_throttle: float = 1.0, hover_throttle: float = -1)

LAV2 vehicle parameters.

Methods:

Name	Description
__post_init__	Calculate derived parameters.

Attributes:

Name	Type	Description
rho	float	空气密度 (kg/m^3)
g	float	重力加速度 (m/s^2)
sim_dt	float	仿真步长 (s)
step_dt	float	控制步长 (s)
randomization	dict[str, dict[str, tuple[float, float]]]	运行时参数随机化配置，按组件 (e.g. rotor/mixer) 划分为多个 scale 相对范围子表
mass	float	质量 (kg)
inertia	list	惯性矩阵 (kg*m^2)
diameter	float	直径 (m)
Cdx	float	空气阻力系数
x	float	x方向臂长 (m)
y	float	y方向臂长 (m)
h	float	重心高度 (m)
alpha	float	y方向涵道外倾角度 (rad)
r_p	float	螺旋桨半径 (m)
J_m	float	电机转动惯量 (kg*m^2)
Ct	float	推力系数
Cq	float	扭矩系数
theta0	float	桨根角 (rad)
thetaTip	float	桨尖角 (rad)
theta1	float	桨叶扭转角 (rad)
lock	float	Lock number
max_rpm	float	最大转速
init_rpm	float	初始转速
rotor_rpm_rate_limit	float	旋翼转速变化率限制 (rpm/s)
cT	float	转速平方到推力增益 (N/(rpm^2))
cM	float	转速平方到扭矩增益 (N*m/(rpm^2))
tau_f	float	电机滤波时间常数 (s)
tau_m	float	兼容保留的电机时间常数 (s)
tau_up	float	电机升速时间常数 (s)
tau_down	float	电机降速时间常数 (s)
r_w	float	主动轮半径 (m)
n_w	int	负重轮个数
B	float	两侧履带中心距 (m)
L	float	履带节点接地部分长度 (m)
h_cg	float	车体质心高度 (m)
f	float	滚动阻力系数(以干燥混凝土路面或沥青路面为例)
mu0	float	滑转率为1时的摩擦系数
K	float	依赖于土壤黏聚系数和摩擦特性的常数
track_cmd_rate_limit	float	履带主动轮指令变化率限制 (rad/s^2)
min_throttle_manual	float	MPC_MANTHR_MIN: Minimum manual thrust (0.0 - 1.0). Default: 0.08
min_throttle	float	MPC_THR_MIN: Minimum thrust in auto thrust control (0.05 - 1.0). Default: 0.12
max_throttle	float	MPC_THR_MAX: Maximum thrust in auto thrust control (0.0 - 1.0). Default: 1.0
hover_throttle	float	MPC_THR_HOVER: Hover throttle manual (0.1 - 0.8). Default: 0.5 in PX4, calculated hover throttle in LAV2

rho class-attribute instance-attribute

rho: float = 1.184

空气密度 (kg/m^3)

g class-attribute instance-attribute

g: float = 9.81

重力加速度 (m/s^2)

sim_dt class-attribute instance-attribute

sim_dt: float = 0.01

仿真步长 (s)

step_dt class-attribute instance-attribute

step_dt: float = 0.01

控制步长 (s)

randomization class-attribute instance-attribute

randomization: dict[str, dict[str, tuple[float, float]]] = field(default_factory=dict)

运行时参数随机化配置，按组件 (e.g. rotor/mixer) 划分为多个 scale 相对范围子表

mass class-attribute instance-attribute

mass: float = 2.1

质量 (kg)

inertia class-attribute instance-attribute

inertia: list = field(default_factory=lambda: tolist())

惯性矩阵 (kg*m^2)

diameter class-attribute instance-attribute

diameter: float = 0.476

直径 (m)

Cdx class-attribute instance-attribute

Cdx: float = 0.5

空气阻力系数

x class-attribute instance-attribute

x: float = 0.0952

x方向臂长 (m)

y class-attribute instance-attribute

y: float = 0.114423

y方向臂长 (m)

h class-attribute instance-attribute

h: float = -0.0125

重心高度 (m)

alpha class-attribute instance-attribute

alpha: float = 8 * (pi / 180)

y方向涵道外倾角度 (rad)

r_p class-attribute instance-attribute

r_p: float = 0.0775

螺旋桨半径 (m)

J_m class-attribute instance-attribute

J_m: float = 1e-05

电机转动惯量 (kg*m^2)

Ct class-attribute instance-attribute

Ct: float = 0.666

推力系数

Cq class-attribute instance-attribute

Cq: float = 0.0716

扭矩系数

theta0 class-attribute instance-attribute

theta0: float = 14.6 * (pi / 180)

桨根角 (rad)

thetaTip class-attribute instance-attribute

thetaTip: float = 6.8 * (pi / 180)

桨尖角 (rad)

theta1 class-attribute instance-attribute

theta1: float = field(init=False)

桨叶扭转角 (rad)

lock class-attribute instance-attribute

lock: float = 0.6051

Lock number

max_rpm class-attribute instance-attribute

max_rpm: float = 10000.0

最大转速

init_rpm class-attribute instance-attribute

init_rpm: float = -1.0

初始转速

rotor_rpm_rate_limit class-attribute instance-attribute

rotor_rpm_rate_limit: float = 50000.0

旋翼转速变化率限制 (rpm/s)

cT class-attribute instance-attribute

cT: float = field(init=False)

转速平方到推力增益 (N/(rpm^2))

cM class-attribute instance-attribute

cM: float = field(init=False)

转速平方到扭矩增益 (N*m/(rpm^2))

tau_f class-attribute instance-attribute

tau_f: float = 0.01

电机滤波时间常数 (s)

tau_m class-attribute instance-attribute

tau_m: float = 0.05

兼容保留的电机时间常数 (s)

tau_up class-attribute instance-attribute

tau_up: float = -1.0

电机升速时间常数 (s)

tau_down class-attribute instance-attribute

tau_down: float = -1.0

电机降速时间常数 (s)

r_w class-attribute instance-attribute

r_w: float = 0.0285

主动轮半径 (m)

n_w class-attribute instance-attribute

n_w: int = 4

负重轮个数

B class-attribute instance-attribute

B: float = 0.136

两侧履带中心距 (m)

L class-attribute instance-attribute

L: float = 0.23

履带节点接地部分长度 (m)

h_cg class-attribute instance-attribute

h_cg: float = 0.078

车体质心高度 (m)

f class-attribute instance-attribute

f: float = 0.01

滚动阻力系数(以干燥混凝土路面或沥青路面为例)

mu0 class-attribute instance-attribute

mu0: float = 0.8

滑转率为1时的摩擦系数

K class-attribute instance-attribute

K: float = 20

依赖于土壤黏聚系数和摩擦特性的常数

track_cmd_rate_limit class-attribute instance-attribute

track_cmd_rate_limit: float = 100

履带主动轮指令变化率限制 (rad/s^2)

min_throttle_manual class-attribute instance-attribute

min_throttle_manual: float = 0.08

MPC_MANTHR_MIN: Minimum manual thrust (0.0 - 1.0). Default: 0.08

min_throttle class-attribute instance-attribute

min_throttle: float = 0.12

MPC_THR_MIN: Minimum thrust in auto thrust control (0.05 - 1.0). Default: 0.12

max_throttle class-attribute instance-attribute

max_throttle: float = 1.0

MPC_THR_MAX: Maximum thrust in auto thrust control (0.0 - 1.0). Default: 1.0

hover_throttle class-attribute instance-attribute

hover_throttle: float = -1

MPC_THR_HOVER: Hover throttle manual (0.1 - 0.8). Default: 0.5 in PX4, calculated hover throttle in LAV2

__post_init__

__post_init__()

Calculate derived parameters.

RotorDynamics

RotorDynamics(params: VehicleParams = VehicleParams(), num_envs: int = 1, device: str | device = 'cpu')

Bases: DynamicsBase

Batched rotor dynamics model.

Initialize batched rotor dynamics on the specified device.

Methods:

Name	Description
randomize	Randomize rotor runtime parameters from nominal values.
update	Calculate the output thrust and torque for each motor based on the motor commands.
reset	Resets the rotor model to initial conditions.

Attributes:

Name	Type	Description
params	VehicleParams	Vehicle parameters for the dynamics model.

params instance-attribute

params: VehicleParams = params

Vehicle parameters for the dynamics model.

randomize

randomize(env_ids: slice | Sequence[int] | Tensor | None = None, randomization: dict[str, tuple[float, float]] | None = None) -> None

Randomize rotor runtime parameters from nominal values.

update

update(commands: Tensor) -> torch.Tensor

Calculate the output thrust and torque for each motor based on the motor commands.

Parameters:

Name	Type	Description	Default
commands	Tensor	Command tensor for each motor. Shape: (num_envs, 4)	required

Returns:

Type	Description
Tensor	torch.Tensor: Thrust and torque tensor for each motor. Shape: (num_envs, 8)

reset

reset(env_ids: slice | Sequence[int] | Tensor) -> None

Resets the rotor model to initial conditions.

Parameters:

Name	Type	Description	Default
env_ids	slice | Sequence[int] | Tensor	The environment ids to reset.	required

ControlActionManager

ControlActionManager(env: GenesisEnv, delay_step: int = 0, params: VehicleParams = VehicleParams(), gains: dict | None = None, limits: dict | None = None, entity_attr: str = 'robot', control_mode: str = 'cmd_motor_speeds')

Bases: BaseActionManager

Base class for action managers that control actuators.

Parameters:

Name	Type	Description	Default
env	GenesisEnv	The environment to manage the DOF actuators for.	required
delay_step	int	The number of steps to delay the actions for. This is an easy way to emulate the latency in the system.	0
params	VehicleParams	Vehicle physical parameters.	VehicleParams()
entity_attr	str	Attribute name on the environment for the robot entity.	'robot'
control_mode	str	Control mode string (e.g. 'cmd_motor_speeds').	'cmd_motor_speeds'

Methods:

Name	Description
step	Handle the received actions.
reset	Reset environments.
get_actions	Get the current actions for the environments.

Attributes:

Name	Type	Description
num_actions	int	The total number of actions.
action_space	tuple[float, float]	If using the default action handler, the action space is [-1, 1].
actions	Tensor	The actions for for the current step.
raw_actions	Tensor	The actions received from the policy, before being converted.

num_actions property

num_actions: int

The total number of actions.

action_space property

action_space: tuple[float, float]

If using the default action handler, the action space is [-1, 1].

actions property

actions: Tensor

The actions for for the current step.

raw_actions property

raw_actions: Tensor

The actions received from the policy, before being converted.

step

step(actions: Tensor) -> None

Handle the received actions.

reset

reset(envs_idx: list[int] | None)

Reset environments.

get_actions

get_actions() -> torch.Tensor

Get the current actions for the environments.

PositionDebugVisualizerConfig

Bases: TypedDict

Defines the configuration for the debug visualizer.

Attributes:

Name	Type	Description
envs_idx	list[int]	The indices of the environments to visualize. If None, all environments will be visualized.
sphere_offset	float	The vertical offset of the debug arrows from the top of the robot
sphere_radius	float	The radius of the shaft of the debug arrows
commanded_color	tuple[float, float, float, float]	The color of the commanded velocity arrow

envs_idx instance-attribute

envs_idx: list[int]

The indices of the environments to visualize. If None, all environments will be visualized.

sphere_offset instance-attribute

sphere_offset: float

The vertical offset of the debug arrows from the top of the robot

sphere_radius instance-attribute

sphere_radius: float

The radius of the shaft of the debug arrows

commanded_color instance-attribute

commanded_color: tuple[float, float, float, float]

The color of the commanded velocity arrow

PositionCommandManager

PositionCommandManager(env: GenesisEnv, range: PositionCommandRange, resample_time_sec: float = 5.0, debug_visualizer: bool = False, debug_visualizer_cfg: PositionDebugVisualizerConfig = DEFAULT_VISUALIZER_CONFIG)

Bases: CommandManager

Generates a position command from uniform distribution. The command comprises of a linear velocity in x and y direction and an angular velocity around the z-axis.

IMPORTANT: The position commands are interpreted as world-relative coordinates: - X-axis: x coordinate of the target position - Y-axis: y coordinate of the target position - Z-axis: z coordinate of the target position

Debug visualization:

If `debug_visualizer` is set to True, a target sphere is rendered above
the commanded position.

Visual meaning:

- GREEN: Commanded position for the robot in the world frame

Parameters:

Name	Type	Description	Default
env	GenesisEnv	The environment to control	required
range	PositionCommandRange	The ranges of linear & angular velocities	required
resample_time_sec	float	The time interval between changing the command	5.0
debug_visualizer	bool	Enable the debug arrow visualization	False
debug_visualizer_cfg	PositionDebugVisualizerConfig	The configuration for the debug visualizer	DEFAULT_VISUALIZER_CONFIG

Example::

class MyEnv(GenesisEnv):
    def config(self):
        # Create a velocity command manager
        self.command_manager = PositionCommandManager(
            self,
            visualize=True,
            range = {
                "lin_vel_x_range": (-1.0, 1.0),
                "lin_vel_y_range": (-1.0, 1.0),
                "ang_vel_z_range": (-0.5, 0.5),
            }
        )

        RewardManager(
            self,
            logging_enabled=True,
            cfg={
                "tracking_lin_vel": {
                    "weight": 1.0,
                    "fn": rewards.command_tracking_lin_vel,
                    "params": {
                        "vel_cmd_manager": self.velocity_command,
                    },
                },
                "tracking_ang_vel": {
                    "weight": 1.0,
                    "fn": rewards.command_tracking_ang_vel,
                    "params": {
                        "vel_cmd_manager": self.velocity_command,
                    },
                },
                # ... other rewards ...
            },
        )

        # Observations
        ObservationManager(
            self,
            cfg={
                "velocity_cmd": {"fn": self.velocity_command.observation},
                # ... other observations ...
            },
        )

Methods:

Name	Description
resample_command	Overwrites commands for environments that should be standing still.
build	Build the position command manager
step	Render the command arrows
use_gamepad	Use a connected gamepad to control the command.

Attributes:

Name	Type	Description
command	Tensor	The desired command value. Shape is (num_envs, num_ranges).

command property

command: Tensor

The desired command value. Shape is (num_envs, num_ranges).

resample_command

resample_command(env_ids: list[int])

Overwrites commands for environments that should be standing still.

build

build()

Build the position command manager

step

step()

Render the command arrows

use_gamepad

use_gamepad(gamepad: Gamepad, pos_x_axis: int = 0, pos_y_axis: int = 1, pos_z_axis: int = 2)

Use a connected gamepad to control the command.

Parameters:

Name	Type	Description	Default
gamepad	Gamepad	The gamepad to use.	required
pos_x_axis	int	Map this gamepad axis index to the position in the x-direction.	0
pos_y_axis	int	Map this gamepad axis index to the position in the y-direction.	1
pos_z_axis	int	Map this gamepad axis index to the position in the z-direction.	2

lin_vel_l2

lin_vel_l2(env: GenesisEnv, entity_attr: str = 'robot', entity_manager: EntityManager = None) -> torch.Tensor

Penalize base linear velocity using L2 squared kernel.

ang_vel_l2

ang_vel_l2(env: GenesisEnv, entity_attr: str = 'robot', entity_manager: EntityManager = None) -> torch.Tensor

Penalize base angular velocity using L2 squared kernel.

pos_error_l2

pos_error_l2(env: GenesisEnv, target_pose: Tensor = None, pose_command: CommandManager = None) -> torch.Tensor

Penalize asset pos from its target pos using L2 squared kernel.

pos_error_tanh

pos_error_tanh(env: GenesisEnv, target_pose: Tensor = None, std: float = 0.8, pose_command: CommandManager = None) -> torch.Tensor

Penalize asset pos from its target pos using tanh kernel.

actions

Classes:

Name	Description
ControlActionManager	Base class for action managers that control actuators.

ControlActionManager

ControlActionManager(env: GenesisEnv, delay_step: int = 0, params: VehicleParams = VehicleParams(), gains: dict | None = None, limits: dict | None = None, entity_attr: str = 'robot', control_mode: str = 'cmd_motor_speeds')

Bases: BaseActionManager

Base class for action managers that control actuators.

Parameters:

Name	Type	Description	Default
env	GenesisEnv	The environment to manage the DOF actuators for.	required
delay_step	int	The number of steps to delay the actions for. This is an easy way to emulate the latency in the system.	0
params	VehicleParams	Vehicle physical parameters.	VehicleParams()
entity_attr	str	Attribute name on the environment for the robot entity.	'robot'
control_mode	str	Control mode string (e.g. 'cmd_motor_speeds').	'cmd_motor_speeds'

Methods:

Name	Description
step	Handle the received actions.
reset	Reset environments.
get_actions	Get the current actions for the environments.

Attributes:

Name	Type	Description
num_actions	int	The total number of actions.
action_space	tuple[float, float]	If using the default action handler, the action space is [-1, 1].
actions	Tensor	The actions for for the current step.
raw_actions	Tensor	The actions received from the policy, before being converted.

num_actions property

num_actions: int

The total number of actions.

action_space property

action_space: tuple[float, float]

If using the default action handler, the action space is [-1, 1].

actions property

actions: Tensor

The actions for for the current step.

raw_actions property

raw_actions: Tensor

The actions received from the policy, before being converted.

step

step(actions: Tensor) -> None

Handle the received actions.

reset

reset(envs_idx: list[int] | None)

Reset environments.

get_actions

get_actions() -> torch.Tensor

Get the current actions for the environments.

commands

Classes:

Name	Description
PositionDebugVisualizerConfig	Defines the configuration for the debug visualizer.
PositionCommandManager	Generates a position command from uniform distribution.

PositionDebugVisualizerConfig

Bases: TypedDict

Defines the configuration for the debug visualizer.

Attributes:

Name	Type	Description
envs_idx	list[int]	The indices of the environments to visualize. If None, all environments will be visualized.
sphere_offset	float	The vertical offset of the debug arrows from the top of the robot
sphere_radius	float	The radius of the shaft of the debug arrows
commanded_color	tuple[float, float, float, float]	The color of the commanded velocity arrow

envs_idx instance-attribute

envs_idx: list[int]

The indices of the environments to visualize. If None, all environments will be visualized.

sphere_offset instance-attribute

sphere_offset: float

The vertical offset of the debug arrows from the top of the robot

sphere_radius instance-attribute

sphere_radius: float

The radius of the shaft of the debug arrows

commanded_color instance-attribute

commanded_color: tuple[float, float, float, float]

The color of the commanded velocity arrow

PositionCommandManager

PositionCommandManager(env: GenesisEnv, range: PositionCommandRange, resample_time_sec: float = 5.0, debug_visualizer: bool = False, debug_visualizer_cfg: PositionDebugVisualizerConfig = DEFAULT_VISUALIZER_CONFIG)

Bases: CommandManager

Generates a position command from uniform distribution. The command comprises of a linear velocity in x and y direction and an angular velocity around the z-axis.

IMPORTANT: The position commands are interpreted as world-relative coordinates: - X-axis: x coordinate of the target position - Y-axis: y coordinate of the target position - Z-axis: z coordinate of the target position

Debug visualization:

If `debug_visualizer` is set to True, a target sphere is rendered above
the commanded position.

Visual meaning:

- GREEN: Commanded position for the robot in the world frame

Parameters:

Name	Type	Description	Default
env	GenesisEnv	The environment to control	required
range	PositionCommandRange	The ranges of linear & angular velocities	required
resample_time_sec	float	The time interval between changing the command	5.0
debug_visualizer	bool	Enable the debug arrow visualization	False
debug_visualizer_cfg	PositionDebugVisualizerConfig	The configuration for the debug visualizer	DEFAULT_VISUALIZER_CONFIG

Example::

class MyEnv(GenesisEnv):
    def config(self):
        # Create a velocity command manager
        self.command_manager = PositionCommandManager(
            self,
            visualize=True,
            range = {
                "lin_vel_x_range": (-1.0, 1.0),
                "lin_vel_y_range": (-1.0, 1.0),
                "ang_vel_z_range": (-0.5, 0.5),
            }
        )

        RewardManager(
            self,
            logging_enabled=True,
            cfg={
                "tracking_lin_vel": {
                    "weight": 1.0,
                    "fn": rewards.command_tracking_lin_vel,
                    "params": {
                        "vel_cmd_manager": self.velocity_command,
                    },
                },
                "tracking_ang_vel": {
                    "weight": 1.0,
                    "fn": rewards.command_tracking_ang_vel,
                    "params": {
                        "vel_cmd_manager": self.velocity_command,
                    },
                },
                # ... other rewards ...
            },
        )

        # Observations
        ObservationManager(
            self,
            cfg={
                "velocity_cmd": {"fn": self.velocity_command.observation},
                # ... other observations ...
            },
        )

Methods:

Name	Description
resample_command	Overwrites commands for environments that should be standing still.
build	Build the position command manager
step	Render the command arrows
use_gamepad	Use a connected gamepad to control the command.

Attributes:

Name	Type	Description
command	Tensor	The desired command value. Shape is (num_envs, num_ranges).

command property

command: Tensor

The desired command value. Shape is (num_envs, num_ranges).

resample_command

resample_command(env_ids: list[int])

Overwrites commands for environments that should be standing still.

build

build()

Build the position command manager

step

step()

Render the command arrows

use_gamepad

use_gamepad(gamepad: Gamepad, pos_x_axis: int = 0, pos_y_axis: int = 1, pos_z_axis: int = 2)

Use a connected gamepad to control the command.

Parameters:

Name	Type	Description	Default
gamepad	Gamepad	The gamepad to use.	required
pos_x_axis	int	Map this gamepad axis index to the position in the x-direction.	0
pos_y_axis	int	Map this gamepad axis index to the position in the y-direction.	1
pos_z_axis	int	Map this gamepad axis index to the position in the z-direction.	2

rewards

Functions:

Name	Description
lin_vel_l2	Penalize base linear velocity using L2 squared kernel.
ang_vel_l2	Penalize base angular velocity using L2 squared kernel.
pos_error_l2	Penalize asset pos from its target pos using L2 squared kernel.
pos_error_tanh	Penalize asset pos from its target pos using tanh kernel.

lin_vel_l2

lin_vel_l2(env: GenesisEnv, entity_attr: str = 'robot', entity_manager: EntityManager = None) -> torch.Tensor

Penalize base linear velocity using L2 squared kernel.

ang_vel_l2

ang_vel_l2(env: GenesisEnv, entity_attr: str = 'robot', entity_manager: EntityManager = None) -> torch.Tensor

Penalize base angular velocity using L2 squared kernel.

pos_error_l2

pos_error_l2(env: GenesisEnv, target_pose: Tensor = None, pose_command: CommandManager = None) -> torch.Tensor

Penalize asset pos from its target pos using L2 squared kernel.

pos_error_tanh

pos_error_tanh(env: GenesisEnv, target_pose: Tensor = None, std: float = 0.8, pose_command: CommandManager = None) -> torch.Tensor

Penalize asset pos from its target pos using tanh kernel.

LAV2_base_velocity

Modules:

Name	Description
environment	LAV2 Base Environment using managers to handle everything.
eval
mdp

environment

LAV2 Base Environment using managers to handle everything.

Classes:

Name	Description
LAV2BaseEnv	Example training environment for the LAV2 robot.

LAV2BaseEnv

LAV2BaseEnv(num_envs: int = 1, dt: float = 1 / 100, max_episode_length_s: int | None = 10, headless: bool = True)

Bases: ManagedEnvironment

Example training environment for the LAV2 robot.

Methods:

Name	Description
config	Configure the environment managers

config

config()

Configure the environment managers

eval

Functions:

Name	Description
get_latest_model	Get the last model from the log directory

get_latest_model

get_latest_model(log_dir: str) -> str

Get the last model from the log directory

mdp

Modules:

Name	Description
actions
rewards

Classes:

Name	Description
FlightActionMapper	Map batched normalized actions into controller targets, wrench, or RPM.
Mixer	Map thrust and moment demands to rotor commands.
VehicleParams	LAV2 vehicle parameters.
RotorDynamics	Batched rotor dynamics model.
ControlActionManager	Base class for action managers that control actuators.

FlightActionMapper

FlightActionMapper(control_mode: str, params: VehicleParams = VehicleParams(), limits: FlightMappingLimits = FlightMappingLimits(), *, num_envs: int = 1, device: str | device = 'cpu')

Map batched normalized actions into controller targets, wrench, or RPM.

Initialize a batched Torch flight action mapper for the selected mode.

Methods:

Name	Description
map_action	Map a normalized batched action into RPM commands and intermediates.

Attributes:

Name	Type	Description
requires_controller	bool	Whether this mode needs a controller target expansion step.

requires_controller property

requires_controller: bool

Whether this mode needs a controller target expansion step.

map_action

map_action(action: Tensor, mixer: Mixer, *, state: Tensor | None = None, flight_controller: ControllerBase | None = None) -> FlightMappingResult

Map a normalized batched action into RPM commands and intermediates.

Mixer

Mixer(params: VehicleParams = VehicleParams(), num_envs: int = 1, device: str | device = 'cpu')

Map thrust and moment demands to rotor commands.

Note

电机推力分配： 电机1 = 推力 - 横滚(roll) + 俯仰(pitch) + 偏航(yaw) 电机2 = 推力 + 横滚(roll) - 俯仰(pitch) + 偏航(yaw) 电机3 = 推力 + 横滚(roll) + 俯仰(pitch) - 偏航(yaw) 电机4 = 推力 - 横滚(roll) - 俯仰(pitch) - 偏航(yaw)

Initialize the torch mixer with vehicle parameters and device.

Methods:

Name	Description
randomize	Randomize mixer runtime parameters from nominal values.
apply_thrust_curve	Apply thrust curve to normalized thrust command.
calculate_rotor_commands	Allocate thrust and moments into rotor speeds.

randomize

randomize(env_ids=None, randomization=None)

Randomize mixer runtime parameters from nominal values.

apply_thrust_curve

apply_thrust_curve(normalized_thrust: Tensor, mode: int = 1) -> torch.Tensor

Apply thrust curve to normalized thrust command.

References

cmd_ctatt: https://github.com/PX4/PX4-Autopilot/blob/e5a483e4b71ab4b994cb0121fe28992d8ce0c8cd/src/modules/mc_att_control/mc_att_control_main.cpp#L111-L137 cmd_ctbr: https://github.com/PX4/PX4-Autopilot/blob/e5a483e4b71ab4b994cb0121fe28992d8ce0c8cd/src/modules/mc_rate_control/MulticopterRateControl.cpp#L156-L179

Parameters:

Name	Type	Description	Default
normalized_thrust	Tensor	Normalized thrust command [-1, 1]. Shape: arbitrary, e.g. (num_envs, n_motors). Applied elementwise.	required
mode	int	Thrust curve mode.	1

Returns:

Type	Description
Tensor	torch.Tensor: Thrust in Newtons with the same shape as normalized_thrust.

Raises:

Type	Description
NotImplementedError	If the specified mode is not implemented.

calculate_rotor_commands

calculate_rotor_commands(control: Tensor) -> torch.Tensor

Allocate thrust and moments into rotor speeds.

Parameters:

Name	Type	Description	Default
control	Tensor	Desired total thrust and roll/pitch/yaw moments. Shape: (num_envs, 4)	required

Returns:

Type	Description
Tensor	torch.Tensor: The amplitude of each motor's rotational speed (calculated using sqrt(clip(rpm_sq))). Shape: (num_envs, 4)

VehicleParams dataclass

VehicleParams(rho: float = 1.184, g: float = 9.81, sim_dt: float = 0.01, step_dt: float = 0.01, randomization: dict[str, dict[str, tuple[float, float]]] = dict(), mass: float = 2.1, inertia: list = (lambda: np.diag([0.11, 0.108, 0.215]).tolist())(), diameter: float = 0.476, Cdx: float = 0.5, x: float = 0.0952, y: float = 0.114423, h: float = -0.0125, alpha: float = 8 * (np.pi / 180), r_p: float = 0.0775, J_m: float = 1e-05, Ct: float = 0.666, Cq: float = 0.0716, theta0: float = 14.6 * (np.pi / 180), thetaTip: float = 6.8 * (np.pi / 180), lock: float = 0.6051, max_rpm: float = 10000.0, init_rpm: float = -1.0, rotor_rpm_rate_limit: float = 50000.0, tau_f: float = 0.01, tau_m: float = 0.05, tau_up: float = -1.0, tau_down: float = -1.0, r_w: float = 0.0285, n_w: int = 4, B: float = 0.136, L: float = 0.23, h_cg: float = 0.078, f: float = 0.01, mu0: float = 0.8, K: float = 20, track_cmd_rate_limit: float = 100, min_throttle_manual: float = 0.08, min_throttle: float = 0.12, max_throttle: float = 1.0, hover_throttle: float = -1)

LAV2 vehicle parameters.

Methods:

Name	Description
__post_init__	Calculate derived parameters.

Attributes:

Name	Type	Description
rho	float	空气密度 (kg/m^3)
g	float	重力加速度 (m/s^2)
sim_dt	float	仿真步长 (s)
step_dt	float	控制步长 (s)
randomization	dict[str, dict[str, tuple[float, float]]]	运行时参数随机化配置，按组件 (e.g. rotor/mixer) 划分为多个 scale 相对范围子表
mass	float	质量 (kg)
inertia	list	惯性矩阵 (kg*m^2)
diameter	float	直径 (m)
Cdx	float	空气阻力系数
x	float	x方向臂长 (m)
y	float	y方向臂长 (m)
h	float	重心高度 (m)
alpha	float	y方向涵道外倾角度 (rad)
r_p	float	螺旋桨半径 (m)
J_m	float	电机转动惯量 (kg*m^2)
Ct	float	推力系数
Cq	float	扭矩系数
theta0	float	桨根角 (rad)
thetaTip	float	桨尖角 (rad)
theta1	float	桨叶扭转角 (rad)
lock	float	Lock number
max_rpm	float	最大转速
init_rpm	float	初始转速
rotor_rpm_rate_limit	float	旋翼转速变化率限制 (rpm/s)
cT	float	转速平方到推力增益 (N/(rpm^2))
cM	float	转速平方到扭矩增益 (N*m/(rpm^2))
tau_f	float	电机滤波时间常数 (s)
tau_m	float	兼容保留的电机时间常数 (s)
tau_up	float	电机升速时间常数 (s)
tau_down	float	电机降速时间常数 (s)
r_w	float	主动轮半径 (m)
n_w	int	负重轮个数
B	float	两侧履带中心距 (m)
L	float	履带节点接地部分长度 (m)
h_cg	float	车体质心高度 (m)
f	float	滚动阻力系数(以干燥混凝土路面或沥青路面为例)
mu0	float	滑转率为1时的摩擦系数
K	float	依赖于土壤黏聚系数和摩擦特性的常数
track_cmd_rate_limit	float	履带主动轮指令变化率限制 (rad/s^2)
min_throttle_manual	float	MPC_MANTHR_MIN: Minimum manual thrust (0.0 - 1.0). Default: 0.08
min_throttle	float	MPC_THR_MIN: Minimum thrust in auto thrust control (0.05 - 1.0). Default: 0.12
max_throttle	float	MPC_THR_MAX: Maximum thrust in auto thrust control (0.0 - 1.0). Default: 1.0
hover_throttle	float	MPC_THR_HOVER: Hover throttle manual (0.1 - 0.8). Default: 0.5 in PX4, calculated hover throttle in LAV2

rho class-attribute instance-attribute

rho: float = 1.184

空气密度 (kg/m^3)

g class-attribute instance-attribute

g: float = 9.81

重力加速度 (m/s^2)

sim_dt class-attribute instance-attribute

sim_dt: float = 0.01

仿真步长 (s)

step_dt class-attribute instance-attribute

step_dt: float = 0.01

控制步长 (s)

randomization class-attribute instance-attribute

randomization: dict[str, dict[str, tuple[float, float]]] = field(default_factory=dict)

运行时参数随机化配置，按组件 (e.g. rotor/mixer) 划分为多个 scale 相对范围子表

mass class-attribute instance-attribute

mass: float = 2.1

质量 (kg)

inertia class-attribute instance-attribute

inertia: list = field(default_factory=lambda: tolist())

惯性矩阵 (kg*m^2)

diameter class-attribute instance-attribute

diameter: float = 0.476

直径 (m)

Cdx class-attribute instance-attribute

Cdx: float = 0.5

空气阻力系数

x class-attribute instance-attribute

x: float = 0.0952

x方向臂长 (m)

y class-attribute instance-attribute

y: float = 0.114423

y方向臂长 (m)

h class-attribute instance-attribute

h: float = -0.0125

重心高度 (m)

alpha class-attribute instance-attribute

alpha: float = 8 * (pi / 180)

y方向涵道外倾角度 (rad)

r_p class-attribute instance-attribute

r_p: float = 0.0775

螺旋桨半径 (m)

J_m class-attribute instance-attribute

J_m: float = 1e-05

电机转动惯量 (kg*m^2)

Ct class-attribute instance-attribute

Ct: float = 0.666

推力系数

Cq class-attribute instance-attribute

Cq: float = 0.0716

扭矩系数

theta0 class-attribute instance-attribute

theta0: float = 14.6 * (pi / 180)

桨根角 (rad)

thetaTip class-attribute instance-attribute

thetaTip: float = 6.8 * (pi / 180)

桨尖角 (rad)

theta1 class-attribute instance-attribute

theta1: float = field(init=False)

桨叶扭转角 (rad)

lock class-attribute instance-attribute

lock: float = 0.6051

Lock number

max_rpm class-attribute instance-attribute

max_rpm: float = 10000.0

最大转速

init_rpm class-attribute instance-attribute

init_rpm: float = -1.0

初始转速

rotor_rpm_rate_limit class-attribute instance-attribute

rotor_rpm_rate_limit: float = 50000.0

旋翼转速变化率限制 (rpm/s)

cT class-attribute instance-attribute

cT: float = field(init=False)

转速平方到推力增益 (N/(rpm^2))

cM class-attribute instance-attribute

cM: float = field(init=False)

转速平方到扭矩增益 (N*m/(rpm^2))

tau_f class-attribute instance-attribute

tau_f: float = 0.01

电机滤波时间常数 (s)

tau_m class-attribute instance-attribute

tau_m: float = 0.05

兼容保留的电机时间常数 (s)

tau_up class-attribute instance-attribute

tau_up: float = -1.0

电机升速时间常数 (s)

tau_down class-attribute instance-attribute

tau_down: float = -1.0

电机降速时间常数 (s)

r_w class-attribute instance-attribute

r_w: float = 0.0285

主动轮半径 (m)

n_w class-attribute instance-attribute

n_w: int = 4

负重轮个数

B class-attribute instance-attribute

B: float = 0.136

两侧履带中心距 (m)

L class-attribute instance-attribute

L: float = 0.23

履带节点接地部分长度 (m)

h_cg class-attribute instance-attribute

h_cg: float = 0.078

车体质心高度 (m)

f class-attribute instance-attribute

f: float = 0.01

滚动阻力系数(以干燥混凝土路面或沥青路面为例)

mu0 class-attribute instance-attribute

mu0: float = 0.8

滑转率为1时的摩擦系数

K class-attribute instance-attribute

K: float = 20

依赖于土壤黏聚系数和摩擦特性的常数

track_cmd_rate_limit class-attribute instance-attribute

track_cmd_rate_limit: float = 100

履带主动轮指令变化率限制 (rad/s^2)

min_throttle_manual class-attribute instance-attribute

min_throttle_manual: float = 0.08

MPC_MANTHR_MIN: Minimum manual thrust (0.0 - 1.0). Default: 0.08

min_throttle class-attribute instance-attribute

min_throttle: float = 0.12

MPC_THR_MIN: Minimum thrust in auto thrust control (0.05 - 1.0). Default: 0.12

max_throttle class-attribute instance-attribute

max_throttle: float = 1.0

MPC_THR_MAX: Maximum thrust in auto thrust control (0.0 - 1.0). Default: 1.0

hover_throttle class-attribute instance-attribute

hover_throttle: float = -1

MPC_THR_HOVER: Hover throttle manual (0.1 - 0.8). Default: 0.5 in PX4, calculated hover throttle in LAV2

__post_init__

__post_init__()

Calculate derived parameters.

RotorDynamics

RotorDynamics(params: VehicleParams = VehicleParams(), num_envs: int = 1, device: str | device = 'cpu')

Bases: DynamicsBase

Batched rotor dynamics model.

Initialize batched rotor dynamics on the specified device.

Methods:

Name	Description
randomize	Randomize rotor runtime parameters from nominal values.
update	Calculate the output thrust and torque for each motor based on the motor commands.
reset	Resets the rotor model to initial conditions.

Attributes:

Name	Type	Description
params	VehicleParams	Vehicle parameters for the dynamics model.

params instance-attribute

params: VehicleParams = params

Vehicle parameters for the dynamics model.

randomize

randomize(env_ids: slice | Sequence[int] | Tensor | None = None, randomization: dict[str, tuple[float, float]] | None = None) -> None

Randomize rotor runtime parameters from nominal values.

update

update(commands: Tensor) -> torch.Tensor

Calculate the output thrust and torque for each motor based on the motor commands.

Parameters:

Name	Type	Description	Default
commands	Tensor	Command tensor for each motor. Shape: (num_envs, 4)	required

Returns:

Type	Description
Tensor	torch.Tensor: Thrust and torque tensor for each motor. Shape: (num_envs, 8)

reset

reset(env_ids: slice | Sequence[int] | Tensor) -> None

Resets the rotor model to initial conditions.

Parameters:

Name	Type	Description	Default
env_ids	slice | Sequence[int] | Tensor	The environment ids to reset.	required

ControlActionManager

ControlActionManager(env: GenesisEnv, delay_step: int = 0, params: VehicleParams = VehicleParams(), entity_attr: str = 'robot')

Bases: BaseActionManager

Base class for action managers that control actuators.

Parameters:

Name	Type	Description	Default
env	GenesisEnv	The environment to manage the DOF actuators for.	required
delay_step	int	The number of steps to delay the actions for. This is an easy way to emulate the latency in the system.	0
params	VehicleParams	Vehicle physical parameters.	VehicleParams()
entity_attr	str	Attribute name on the environment for the robot entity.	'robot'

Methods:

Name	Description
step	Handle the received actions.
reset	Reset environments.
get_actions	Get the current actions for the environments.

Attributes:

Name	Type	Description
num_actions	int	The total number of actions.
action_space	tuple[float, float]	If using the default action handler, the action space is [-1, 1].
actions	Tensor	The actions for for the current step.
raw_actions	Tensor	The actions received from the policy, before being converted.

num_actions property

num_actions: int

The total number of actions.

action_space property

action_space: tuple[float, float]

If using the default action handler, the action space is [-1, 1].

actions property

actions: Tensor

The actions for for the current step.

raw_actions property

raw_actions: Tensor

The actions received from the policy, before being converted.

step

step(actions: Tensor) -> None

Handle the received actions.

reset

reset(envs_idx: list[int] | None)

Reset environments.

get_actions

get_actions() -> torch.Tensor

Get the current actions for the environments.

actions

Classes:

Name	Description
ControlActionManager	Base class for action managers that control actuators.

ControlActionManager

ControlActionManager(env: GenesisEnv, delay_step: int = 0, params: VehicleParams = VehicleParams(), entity_attr: str = 'robot')

Bases: BaseActionManager

Base class for action managers that control actuators.

Parameters:

Name	Type	Description	Default
env	GenesisEnv	The environment to manage the DOF actuators for.	required
delay_step	int	The number of steps to delay the actions for. This is an easy way to emulate the latency in the system.	0
params	VehicleParams	Vehicle physical parameters.	VehicleParams()
entity_attr	str	Attribute name on the environment for the robot entity.	'robot'

Methods:

Name	Description
step	Handle the received actions.
reset	Reset environments.
get_actions	Get the current actions for the environments.

Attributes:

Name	Type	Description
num_actions	int	The total number of actions.
action_space	tuple[float, float]	If using the default action handler, the action space is [-1, 1].
actions	Tensor	The actions for for the current step.
raw_actions	Tensor	The actions received from the policy, before being converted.

num_actions property

num_actions: int

The total number of actions.

action_space property

action_space: tuple[float, float]

If using the default action handler, the action space is [-1, 1].

actions property

actions: Tensor

The actions for for the current step.

raw_actions property

raw_actions: Tensor

The actions received from the policy, before being converted.

step

step(actions: Tensor) -> None

Handle the received actions.

reset

reset(envs_idx: list[int] | None)

Reset environments.

get_actions

get_actions() -> torch.Tensor

Get the current actions for the environments.

rewards

Functions:

Name	Description
lin_vel_l2	Penalize base linear velocity using L2 squared kernel.
ang_vel_l2	Penalize base angular velocity using L2 squared kernel.
pos_error_l2	Penalize asset pos from its target pos using L2 squared kernel.
pos_error_tanh	Penalize asset pos from its target pos using tanh kernel.
yaw_error_l2	Penalize heading error from target heading using L2 squared kernel.
yaw_error_tanh	Penalize heading error from target heading.
wrap_to_pi	Wraps input angles (in radians) to the range :math:[-\pi, \pi].

lin_vel_l2

lin_vel_l2(env: GenesisEnv, entity_attr: str = 'robot', entity_manager: EntityManager = None) -> torch.Tensor

Penalize base linear velocity using L2 squared kernel.

ang_vel_l2

ang_vel_l2(env: GenesisEnv, entity_attr: str = 'robot', entity_manager: EntityManager = None) -> torch.Tensor

Penalize base angular velocity using L2 squared kernel.

pos_error_l2

pos_error_l2(env: GenesisEnv, target_pose: Tensor, pose_command: CommandManager = None, entity_attr: str = 'robot', entity_manager: EntityManager = None) -> torch.Tensor

Penalize asset pos from its target pos using L2 squared kernel.

pos_error_tanh

pos_error_tanh(env: GenesisEnv, target_pose: Tensor, pose_command: CommandManager = None, entity_attr: str = 'robot', entity_manager: EntityManager = None) -> torch.Tensor

Penalize asset pos from its target pos using tanh kernel.

yaw_error_l2

yaw_error_l2(env: GenesisEnv, target_pose: Tensor, pose_command: CommandManager = None, entity_attr: str = 'robot', entity_manager: EntityManager = None) -> torch.Tensor

Penalize heading error from target heading using L2 squared kernel.

yaw_error_tanh

yaw_error_tanh(env: GenesisEnv, target_pose: Tensor, pose_command: CommandManager = None, entity_attr: str = 'robot', entity_manager: EntityManager = None, std: float = 0.8) -> torch.Tensor

Penalize heading error from target heading.

wrap_to_pi

wrap_to_pi(angles: Tensor) -> torch.Tensor

Wraps input angles (in radians) to the range :math:[-\pi, \pi].

This function wraps angles in radians to the range :math:[-\pi, \pi], such that :math:\pi maps to :math:\pi, and :math:-\pi maps to :math:-\pi. In general, odd positive multiples of :math:\pi are mapped to :math:\pi, and odd negative multiples of :math:\pi are mapped to :math:-\pi.

The function behaves similar to MATLAB's wrapToPi <https://www.mathworks.com/help/map/ref/wraptopi.html>_ function.

Parameters:

Name	Type	Description	Default
angles	Tensor	Input angles of any shape.	required

Returns:

Type	Description
Tensor	torch.Tensor: Angles in the range :math:[-\pi, \pi].

isaaclab

Modules:

Name	Description
LAV2_base
LAV2_base_direct
LAV2_navrl
LAV2_trajectory

LAV2_base

Modules:

Name	Description
LAV2_bimodal_vel_env_cfg
LAV2_env_cfg
LAV2_track_env_cfg
LAV2_vel_env_cfg
mdp

LAV2_bimodal_vel_env_cfg

Classes:

Name	Description
ActionsCfg	Action specifications for the MDP.
ObservationsCfg	Observation specifications for the MDP.
EventCfg	Configuration for events.
CommandsCfg	Command specifications for the MDP.
RewardsCfg	Reward terms for the MDP.
TerminationsCfg	Termination terms for the MDP.
LAV2EnvCfg	Configuration for the LAV2 environment.

ActionsCfg

Action specifications for the MDP.

ObservationsCfg

Observation specifications for the MDP.

Classes:

Name	Description
PolicyCfg	Observations for policy group.

PolicyCfg

Bases: ObservationGroupCfg

Observations for policy group.

EventCfg

Configuration for events.

CommandsCfg

Command specifications for the MDP.

RewardsCfg

Reward terms for the MDP.

TerminationsCfg

Termination terms for the MDP.

LAV2EnvCfg

Bases: ManagerBasedRLEnvCfg

Configuration for the LAV2 environment.

Methods:

Name	Description
__post_init__	Post initialization.

__post_init__

__post_init__() -> None

Post initialization.

LAV2_env_cfg

Classes:

Name	Description
ActionsCfg	Action specifications for the MDP.
ObservationsCfg	Observation specifications for the MDP.
EventCfg	Configuration for events.
CommandsCfg	Command specifications for the MDP.
RewardsCfg	Reward terms for the MDP.
TerminationsCfg	Termination terms for the MDP.
LAV2EnvCfg	Configuration for the LAV2 environment.

ActionsCfg

Action specifications for the MDP.

ObservationsCfg

Observation specifications for the MDP.

Classes:

Name	Description
PolicyCfg	Observations for policy group.

PolicyCfg

Bases: ObservationGroupCfg

Observations for policy group.

EventCfg

Configuration for events.

CommandsCfg

Command specifications for the MDP.

RewardsCfg

Reward terms for the MDP.

TerminationsCfg

Termination terms for the MDP.

LAV2EnvCfg

Bases: ManagerBasedRLEnvCfg

Configuration for the LAV2 environment.

Methods:

Name	Description
__post_init__	Post initialization.

__post_init__

__post_init__() -> None

Post initialization.

LAV2_track_env_cfg

Classes:

Name	Description
ActionsCfg	Action specifications for the MDP.
ObservationsCfg	Observation specifications for the MDP.
EventCfg	Configuration for events.
CommandsCfg	Command specifications for the MDP.
RewardsCfg	Reward terms for the MDP.
TerminationsCfg	Termination terms for the MDP.
LAV2EnvCfg	Configuration for the LAV2 environment.

ActionsCfg

Action specifications for the MDP.

ObservationsCfg

Observation specifications for the MDP.

Classes:

Name	Description
PolicyCfg	Observations for policy group.

PolicyCfg

Bases: ObservationGroupCfg

Observations for policy group.

EventCfg

Configuration for events.

CommandsCfg

Command specifications for the MDP.

RewardsCfg

Reward terms for the MDP.

TerminationsCfg

Termination terms for the MDP.

LAV2EnvCfg

Bases: ManagerBasedRLEnvCfg

Configuration for the LAV2 environment.

Methods:

Name	Description
__post_init__	Post initialization.

__post_init__

__post_init__() -> None

Post initialization.

LAV2_vel_env_cfg

Classes:

Name	Description
ActionsCfg	Action specifications for the MDP.
ObservationsCfg	Observation specifications for the MDP.
EventCfg	Configuration for events.
CommandsCfg	Command specifications for the MDP.
RewardsCfg	Reward terms for the MDP.
TerminationsCfg	Termination terms for the MDP.
LAV2EnvCfg	Configuration for the LAV2 environment.

ActionsCfg

Action specifications for the MDP.

ObservationsCfg

Observation specifications for the MDP.

Classes:

Name	Description
PolicyCfg	Observations for policy group.

PolicyCfg

Bases: ObservationGroupCfg

Observations for policy group.

EventCfg

Configuration for events.

CommandsCfg

Command specifications for the MDP.

RewardsCfg

Reward terms for the MDP.

TerminationsCfg

Termination terms for the MDP.

LAV2EnvCfg

Bases: ManagerBasedRLEnvCfg

Configuration for the LAV2 environment.

Methods:

Name	Description
__post_init__	Post initialization.

__post_init__

__post_init__() -> None

Post initialization.

mdp

Modules:

Name	Description
actions
commands
events
observations
rewards
terminations

Classes:

Name	Description
FlightActionMapper	Map batched normalized actions into controller targets, wrench, or RPM.
FlightMappingLimits	Per-channel limits used when mapping normalized actions to targets.
Mixer	Map thrust and moment demands to rotor commands.
FlightController	PID flight controller (batched).
VehicleParams	LAV2 vehicle parameters.
RotorDynamics	Batched rotor dynamics model.
TrackDynamics	Batched track dynamics model.
ControlAction	Body torque control action term.
ControlActionCfg	See :class:ControlAction for more details.
TrackControlAction	Track force control action term.
TrackControlActionCfg	See :class:ControlAction for more details.
UniformPoseCommandGlobal	Command generator that generates pose commands containing a 3-D position and heading.
UniformPoseCommandGlobalCfg	Configuration for the uniform 3D-pose command generator.
UniformVelocityGlobalCommand	Command generator that generates a velocity command with uniform distribution.
UniformVelocityGlobalCommandCfg	Configuration for the uniform velocity command generator.
UniformBimodalVelocityCommand	Bimodal velocity command generator with a discrete mode indicator (4+1).
UniformBimodalVelocityCommandCfg	Configuration for the bimodal velocity command generator (flight vs ground).

Functions:

Name	Description
randomize_action_term_attr	Randomize a selected action-term attribute.
randomize_controller_params	Randomize controller gains and command limits attached to an action term.
randomize_mixer_params	Randomize mixer parameters attached to an action term.
randomize_rotor_params	Randomize rotor parameters attached to an action term.
target_pos_b	Position of target in body frame.
root_euler_w	Euler angles of the root in world frame.
lin_vel_l2	Penalize base linear velocity using L2 squared kernel.
ang_vel_l2	Penalize base angular velocity using L2 squared kernel.
pos_error_l2	Penalize asset pos from its target pos using L2 squared kernel.
pos_error_tanh	Penalize asset pos from its target pos using tanh kernel.
yaw_error_l2	Penalize heading error from target heading using L2 squared kernel.
yaw_error_tanh	Penalize heading error from target heading using tanh kernel.
track_lin_vel_z_exp	Reward tracking of linear velocity commands (z axis) using exponential kernel.
track_lin_vel_exp	Reward tracking of linear velocity commands using exponential kernel.
track_yaw_vel_exp	Reward tracking of angular velocity commands (yaw) using exponential kernel.
bimodal_action_tanh	Penalize bimodal actions using tanh kernel.
bimodal_height_tanh	Penalize bimodal height using tanh kernel.
contact_impulse	Penalize excessive contact impulse (rate of change of contact forces).
bimodal_contacts	Penalize contacts when switching from flight to ground mode.
died	Termination condition based on the robot's height.

FlightActionMapper

FlightActionMapper(control_mode: str, params: VehicleParams = VehicleParams(), limits: FlightMappingLimits = FlightMappingLimits(), *, num_envs: int = 1, device: str | device = 'cpu')

Map batched normalized actions into controller targets, wrench, or RPM.

Initialize a batched Torch flight action mapper for the selected mode.

Methods:

Name	Description
map_action	Map a normalized batched action into RPM commands and intermediates.

Attributes:

Name	Type	Description
requires_controller	bool	Whether this mode needs a controller target expansion step.

requires_controller property

requires_controller: bool

Whether this mode needs a controller target expansion step.

map_action

map_action(action: Tensor, mixer: Mixer, *, state: Tensor | None = None, flight_controller: ControllerBase | None = None) -> FlightMappingResult

Map a normalized batched action into RPM commands and intermediates.

FlightMappingLimits dataclass

FlightMappingLimits(vel_limits: tuple[float, float, float] = (5.0, 5.0, 5.0), att_limits: tuple[float, float, float] = (1.0, 1.0, math.pi), br_limits: tuple[float, float, float] = (2.0, 2.0, 2.0))

Per-channel limits used when mapping normalized actions to targets.

Mixer

Mixer(params: VehicleParams = VehicleParams(), num_envs: int = 1, device: str | device = 'cpu')

Map thrust and moment demands to rotor commands.

Note

电机推力分配： 电机1 = 推力 - 横滚(roll) + 俯仰(pitch) + 偏航(yaw) 电机2 = 推力 + 横滚(roll) - 俯仰(pitch) + 偏航(yaw) 电机3 = 推力 + 横滚(roll) + 俯仰(pitch) - 偏航(yaw) 电机4 = 推力 - 横滚(roll) - 俯仰(pitch) - 偏航(yaw)

Initialize the torch mixer with vehicle parameters and device.

Methods:

Name	Description
randomize	Randomize mixer runtime parameters from nominal values.
apply_thrust_curve	Apply thrust curve to normalized thrust command.
calculate_rotor_commands	Allocate thrust and moments into rotor speeds.

randomize

randomize(env_ids=None, randomization=None)

Randomize mixer runtime parameters from nominal values.

apply_thrust_curve

apply_thrust_curve(normalized_thrust: Tensor, mode: int = 1) -> torch.Tensor

Apply thrust curve to normalized thrust command.

References

cmd_ctatt: https://github.com/PX4/PX4-Autopilot/blob/e5a483e4b71ab4b994cb0121fe28992d8ce0c8cd/src/modules/mc_att_control/mc_att_control_main.cpp#L111-L137 cmd_ctbr: https://github.com/PX4/PX4-Autopilot/blob/e5a483e4b71ab4b994cb0121fe28992d8ce0c8cd/src/modules/mc_rate_control/MulticopterRateControl.cpp#L156-L179

Parameters:

Name	Type	Description	Default
normalized_thrust	Tensor	Normalized thrust command [-1, 1]. Shape: arbitrary, e.g. (num_envs, n_motors). Applied elementwise.	required
mode	int	Thrust curve mode.	1

Returns:

Type	Description
Tensor	torch.Tensor: Thrust in Newtons with the same shape as normalized_thrust.

Raises:

Type	Description
NotImplementedError	If the specified mode is not implemented.

calculate_rotor_commands

calculate_rotor_commands(control: Tensor) -> torch.Tensor

Allocate thrust and moments into rotor speeds.

Parameters:

Name	Type	Description	Default
control	Tensor	Desired total thrust and roll/pitch/yaw moments. Shape: (num_envs, 4)	required

Returns:

Type	Description
Tensor	torch.Tensor: The amplitude of each motor's rotational speed (calculated using sqrt(clip(rpm_sq))). Shape: (num_envs, 4)

FlightController

FlightController(params: VehicleParams = VehicleParams(), control_mask: list[int] = [1, 1, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0], gains: dict | None = None, limits: dict | None = None, num_envs: int = 1, device: str | device = 'cpu')

Bases: ControllerBase

PID flight controller (batched).

Initialize the batched flight PID controller.

Methods:

Name	Description
randomize	Randomize configured PID gains and command limits for selected environments.
reset	Reset PID loops for selected environments.
update	Update controller.

Attributes:

Name	Type	Description
params	VehicleParams	Vehicle parameters for the controller.
control_mask	ndarray | Tensor	unified control mask for target/state channels, used in update(). e.g. using 9~11 for attitude yaw control.

params instance-attribute

params: VehicleParams = params

Vehicle parameters for the controller.

control_mask instance-attribute

control_mask: ndarray | Tensor = control_mask

unified control mask for target/state channels, used in update(). e.g. using 9~11 for attitude yaw control.

randomize

randomize(env_ids=None, gains=None, limits=None)

Randomize configured PID gains and command limits for selected environments.

reset

reset(env_ids)

Reset PID loops for selected environments.

update

update(target: Tensor, state: Tensor) -> torch.Tensor

Update controller.

Parameters:

Name	Type	Description	Default
target	Tensor	(N, 12) desired [x, y, z, vx, vy, vz, roll, pitch, yaw, wx, wy, wz]	required
state	Tensor	(N, 12) current state same ordering.	required

Returns:

Type	Description
Tensor	torch.Tensor: Tensor of shape (N, 4): [thrust, roll_moment, pitch_moment, yaw_moment].

VehicleParams dataclass

VehicleParams(rho: float = 1.184, g: float = 9.81, sim_dt: float = 0.01, step_dt: float = 0.01, randomization: dict[str, dict[str, tuple[float, float]]] = dict(), mass: float = 2.1, inertia: list = (lambda: np.diag([0.11, 0.108, 0.215]).tolist())(), diameter: float = 0.476, Cdx: float = 0.5, x: float = 0.0952, y: float = 0.114423, h: float = -0.0125, alpha: float = 8 * (np.pi / 180), r_p: float = 0.0775, J_m: float = 1e-05, Ct: float = 0.666, Cq: float = 0.0716, theta0: float = 14.6 * (np.pi / 180), thetaTip: float = 6.8 * (np.pi / 180), lock: float = 0.6051, max_rpm: float = 10000.0, init_rpm: float = -1.0, rotor_rpm_rate_limit: float = 50000.0, tau_f: float = 0.01, tau_m: float = 0.05, tau_up: float = -1.0, tau_down: float = -1.0, r_w: float = 0.0285, n_w: int = 4, B: float = 0.136, L: float = 0.23, h_cg: float = 0.078, f: float = 0.01, mu0: float = 0.8, K: float = 20, track_cmd_rate_limit: float = 100, min_throttle_manual: float = 0.08, min_throttle: float = 0.12, max_throttle: float = 1.0, hover_throttle: float = -1)

LAV2 vehicle parameters.

Methods:

Name	Description
__post_init__	Calculate derived parameters.

Attributes:

Name	Type	Description
rho	float	空气密度 (kg/m^3)
g	float	重力加速度 (m/s^2)
sim_dt	float	仿真步长 (s)
step_dt	float	控制步长 (s)
randomization	dict[str, dict[str, tuple[float, float]]]	运行时参数随机化配置，按组件 (e.g. rotor/mixer) 划分为多个 scale 相对范围子表
mass	float	质量 (kg)
inertia	list	惯性矩阵 (kg*m^2)
diameter	float	直径 (m)
Cdx	float	空气阻力系数
x	float	x方向臂长 (m)
y	float	y方向臂长 (m)
h	float	重心高度 (m)
alpha	float	y方向涵道外倾角度 (rad)
r_p	float	螺旋桨半径 (m)
J_m	float	电机转动惯量 (kg*m^2)
Ct	float	推力系数
Cq	float	扭矩系数
theta0	float	桨根角 (rad)
thetaTip	float	桨尖角 (rad)
theta1	float	桨叶扭转角 (rad)
lock	float	Lock number
max_rpm	float	最大转速
init_rpm	float	初始转速
rotor_rpm_rate_limit	float	旋翼转速变化率限制 (rpm/s)
cT	float	转速平方到推力增益 (N/(rpm^2))
cM	float	转速平方到扭矩增益 (N*m/(rpm^2))
tau_f	float	电机滤波时间常数 (s)
tau_m	float	兼容保留的电机时间常数 (s)
tau_up	float	电机升速时间常数 (s)
tau_down	float	电机降速时间常数 (s)
r_w	float	主动轮半径 (m)
n_w	int	负重轮个数
B	float	两侧履带中心距 (m)
L	float	履带节点接地部分长度 (m)
h_cg	float	车体质心高度 (m)
f	float	滚动阻力系数(以干燥混凝土路面或沥青路面为例)
mu0	float	滑转率为1时的摩擦系数
K	float	依赖于土壤黏聚系数和摩擦特性的常数
track_cmd_rate_limit	float	履带主动轮指令变化率限制 (rad/s^2)
min_throttle_manual	float	MPC_MANTHR_MIN: Minimum manual thrust (0.0 - 1.0). Default: 0.08
min_throttle	float	MPC_THR_MIN: Minimum thrust in auto thrust control (0.05 - 1.0). Default: 0.12
max_throttle	float	MPC_THR_MAX: Maximum thrust in auto thrust control (0.0 - 1.0). Default: 1.0
hover_throttle	float	MPC_THR_HOVER: Hover throttle manual (0.1 - 0.8). Default: 0.5 in PX4, calculated hover throttle in LAV2

rho class-attribute instance-attribute

rho: float = 1.184

空气密度 (kg/m^3)

g class-attribute instance-attribute

g: float = 9.81

重力加速度 (m/s^2)

sim_dt class-attribute instance-attribute

sim_dt: float = 0.01

仿真步长 (s)

step_dt class-attribute instance-attribute

step_dt: float = 0.01

控制步长 (s)

randomization class-attribute instance-attribute

randomization: dict[str, dict[str, tuple[float, float]]] = field(default_factory=dict)

运行时参数随机化配置，按组件 (e.g. rotor/mixer) 划分为多个 scale 相对范围子表

mass class-attribute instance-attribute

mass: float = 2.1

质量 (kg)

inertia class-attribute instance-attribute

inertia: list = field(default_factory=lambda: tolist())

惯性矩阵 (kg*m^2)

diameter class-attribute instance-attribute

diameter: float = 0.476

直径 (m)

Cdx class-attribute instance-attribute

Cdx: float = 0.5

空气阻力系数

x class-attribute instance-attribute

x: float = 0.0952

x方向臂长 (m)

y class-attribute instance-attribute

y: float = 0.114423

y方向臂长 (m)

h class-attribute instance-attribute

h: float = -0.0125

重心高度 (m)

alpha class-attribute instance-attribute

alpha: float = 8 * (pi / 180)

y方向涵道外倾角度 (rad)

r_p class-attribute instance-attribute

r_p: float = 0.0775

螺旋桨半径 (m)

J_m class-attribute instance-attribute

J_m: float = 1e-05

电机转动惯量 (kg*m^2)

Ct class-attribute instance-attribute

Ct: float = 0.666

推力系数

Cq class-attribute instance-attribute

Cq: float = 0.0716

扭矩系数

theta0 class-attribute instance-attribute

theta0: float = 14.6 * (pi / 180)

桨根角 (rad)

thetaTip class-attribute instance-attribute

thetaTip: float = 6.8 * (pi / 180)

桨尖角 (rad)

theta1 class-attribute instance-attribute

theta1: float = field(init=False)

桨叶扭转角 (rad)

lock class-attribute instance-attribute

lock: float = 0.6051

Lock number

max_rpm class-attribute instance-attribute

max_rpm: float = 10000.0

最大转速

init_rpm class-attribute instance-attribute

init_rpm: float = -1.0

初始转速

rotor_rpm_rate_limit class-attribute instance-attribute

rotor_rpm_rate_limit: float = 50000.0

旋翼转速变化率限制 (rpm/s)

cT class-attribute instance-attribute

cT: float = field(init=False)

转速平方到推力增益 (N/(rpm^2))

cM class-attribute instance-attribute

cM: float = field(init=False)

转速平方到扭矩增益 (N*m/(rpm^2))

tau_f class-attribute instance-attribute

tau_f: float = 0.01

电机滤波时间常数 (s)

tau_m class-attribute instance-attribute

tau_m: float = 0.05

兼容保留的电机时间常数 (s)

tau_up class-attribute instance-attribute

tau_up: float = -1.0

电机升速时间常数 (s)

tau_down class-attribute instance-attribute

tau_down: float = -1.0

电机降速时间常数 (s)

r_w class-attribute instance-attribute

r_w: float = 0.0285

主动轮半径 (m)

n_w class-attribute instance-attribute

n_w: int = 4

负重轮个数

B class-attribute instance-attribute

B: float = 0.136

两侧履带中心距 (m)

L class-attribute instance-attribute

L: float = 0.23

履带节点接地部分长度 (m)

h_cg class-attribute instance-attribute

h_cg: float = 0.078

车体质心高度 (m)

f class-attribute instance-attribute

f: float = 0.01

滚动阻力系数(以干燥混凝土路面或沥青路面为例)

mu0 class-attribute instance-attribute

mu0: float = 0.8

滑转率为1时的摩擦系数

K class-attribute instance-attribute

K: float = 20

依赖于土壤黏聚系数和摩擦特性的常数

track_cmd_rate_limit class-attribute instance-attribute

track_cmd_rate_limit: float = 100

履带主动轮指令变化率限制 (rad/s^2)

min_throttle_manual class-attribute instance-attribute

min_throttle_manual: float = 0.08

MPC_MANTHR_MIN: Minimum manual thrust (0.0 - 1.0). Default: 0.08

min_throttle class-attribute instance-attribute

min_throttle: float = 0.12

MPC_THR_MIN: Minimum thrust in auto thrust control (0.05 - 1.0). Default: 0.12

max_throttle class-attribute instance-attribute

max_throttle: float = 1.0

MPC_THR_MAX: Maximum thrust in auto thrust control (0.0 - 1.0). Default: 1.0

hover_throttle class-attribute instance-attribute

hover_throttle: float = -1

MPC_THR_HOVER: Hover throttle manual (0.1 - 0.8). Default: 0.5 in PX4, calculated hover throttle in LAV2

__post_init__

__post_init__()

Calculate derived parameters.

RotorDynamics

RotorDynamics(params: VehicleParams = VehicleParams(), num_envs: int = 1, device: str | device = 'cpu')

Bases: DynamicsBase

Batched rotor dynamics model.

Initialize batched rotor dynamics on the specified device.

Methods:

Name	Description
randomize	Randomize rotor runtime parameters from nominal values.
update	Calculate the output thrust and torque for each motor based on the motor commands.
reset	Resets the rotor model to initial conditions.

Attributes:

Name	Type	Description
params	VehicleParams	Vehicle parameters for the dynamics model.

params instance-attribute

params: VehicleParams = params

Vehicle parameters for the dynamics model.

randomize

randomize(env_ids: slice | Sequence[int] | Tensor | None = None, randomization: dict[str, tuple[float, float]] | None = None) -> None

Randomize rotor runtime parameters from nominal values.

update

update(commands: Tensor) -> torch.Tensor

Calculate the output thrust and torque for each motor based on the motor commands.

Parameters:

Name	Type	Description	Default
commands	Tensor	Command tensor for each motor. Shape: (num_envs, 4)	required

Returns:

Type	Description
Tensor	torch.Tensor: Thrust and torque tensor for each motor. Shape: (num_envs, 8)

reset

reset(env_ids: slice | Sequence[int] | Tensor) -> None

Resets the rotor model to initial conditions.

Parameters:

Name	Type	Description	Default
env_ids	slice | Sequence[int] | Tensor	The environment ids to reset.	required

TrackDynamics

TrackDynamics(params: VehicleParams = VehicleParams(), num_envs: int = 1, device: str | device = 'cpu')

Bases: DynamicsBase

Batched track dynamics model.

Initialize batched track dynamics on the specified device.

Methods:

Name	Description
update	Compute track forces for batched environments.
reset	Resets the track model to initial conditions.

Attributes:

Name	Type	Description
params	VehicleParams	Vehicle parameters for the dynamics model.

params instance-attribute

params: VehicleParams = params

Vehicle parameters for the dynamics model.

update

update(commands: Tensor, state: Tensor) -> torch.Tensor

Compute track forces for batched environments.

Parameters:

Name	Type	Description	Default
commands	Tensor	Shape (num_envs, 2) track commands [left, right].	required
state	Tensor	Shape (num_envs, 3) body-frame velocities [u, v, r].	required

Returns:

Type	Description
Tensor	torch.Tensor: Tensor of shape (num_envs, 2, n_w, 2) [side, wheel, (Fx, Fy)].

reset

reset(env_ids: slice | Sequence[int] | Tensor) -> None

Resets the track model to initial conditions.

Parameters:

Name	Type	Description	Default
env_ids	slice | Sequence[int] | Tensor	The environment ids to reset.	required

ControlAction

ControlAction(cfg: ControlActionCfg, env: ManagerBasedRLEnv)

Bases: ActionTerm

Body torque control action term.

This action term applies a wrench to the drone body frame based on action commands

Attributes:

Name	Type	Description
cfg	ControlActionCfg	The configuration of the action term.

cfg instance-attribute

cfg: ControlActionCfg = cfg

The configuration of the action term.

ControlActionCfg

Bases: ActionTermCfg

See :class:ControlAction for more details.

Attributes:

Name	Type	Description
class_type	type[ActionTerm]	Class of the action term.
asset_name	str	Name of the asset in the environment for which the commands are generated.
params	VehicleParams	Vehicle parameters.
gains	dict | None	Optional gains configuration for the flight controller. See FlightController for details.
limits	dict | None	Optional command-limit configuration passed directly to the flight controller.
control_mode	str	Control mode, options: cmd_motor_thrusts, cmd_motor_speeds, cmd_ctbm, cmd_ctbr, cmd_ctatt, cmd_vel, cmd_pos.

class_type class-attribute instance-attribute

class_type: type[ActionTerm] = ControlAction

Class of the action term.

asset_name class-attribute instance-attribute

asset_name: str = 'robot'

Name of the asset in the environment for which the commands are generated.

params class-attribute instance-attribute

params: VehicleParams = VehicleParams()

Vehicle parameters.

gains class-attribute instance-attribute

gains: dict | None = None

Optional gains configuration for the flight controller. See FlightController for details.

limits class-attribute instance-attribute

limits: dict | None = None

Optional command-limit configuration passed directly to the flight controller.

control_mode class-attribute instance-attribute

control_mode: str = 'cmd_motor_speeds'

Control mode, options: cmd_motor_thrusts, cmd_motor_speeds, cmd_ctbm, cmd_ctbr, cmd_ctatt, cmd_vel, cmd_pos.

rotorpy ref: - https://github.com/spencerfolk/rotorpy/blob/main/rotorpy/learning/quadrotor_environments.py#L319C1-L350C12 - https://github.com/spencerfolk/rotorpy/blob/main/rotorpy/vehicles/multirotor.py#L997

TODO: also align control modes with px4, see: https://docs.px4.io/v1.14/en/flight_modes/offboard (version is unclear ATM)

TrackControlAction

TrackControlAction(cfg: TrackControlActionCfg, env: ManagerBasedRLEnv)

Bases: ActionTerm

Track force control action term.

This action term converts normalized track commands into ground reaction forces applied on each track body.

Attributes:

Name	Type	Description
cfg	TrackControlActionCfg	The configuration of the action term.

cfg instance-attribute

cfg: TrackControlActionCfg = cfg

The configuration of the action term.

TrackControlActionCfg

Bases: ActionTermCfg

See :class:ControlAction for more details.

Attributes:

Name	Type	Description
class_type	type[ActionTerm]	Class of the action term.
asset_name	str	Name of the asset in the environment for which the commands are generated.
params	VehicleParams	Vehicle parameters.

class_type class-attribute instance-attribute

class_type: type[ActionTerm] = TrackControlAction

Class of the action term.

asset_name class-attribute instance-attribute

asset_name: str = 'robot'

Name of the asset in the environment for which the commands are generated.

params class-attribute instance-attribute

params: VehicleParams = VehicleParams()

Vehicle parameters.

UniformPoseCommandGlobal

UniformPoseCommandGlobal(cfg: UniformPoseCommandGlobalCfg, env: ManagerBasedEnv)

Bases: CommandTerm

Command generator that generates pose commands containing a 3-D position and heading.

The command generator samples uniform 3D positions around the environment origin. The heading command is either set to point towards the target or is sampled uniformly. This can be configured through the :attr:UniformPoseCommandGlobalCfg.simple_heading parameter in the configuration.

Initialize the command generator class.

Parameters:

Name	Type	Description	Default
cfg	UniformPoseCommandGlobalCfg	The configuration parameters for the command generator.	required
env	ManagerBasedEnv	The environment object.	required

Attributes:

Name	Type	Description
cfg	UniformPoseCommandGlobalCfg	Configuration for the command generator.
command	Tensor	The desired 3D-pose in base frame. Shape is (num_envs, 4).

cfg instance-attribute

cfg: UniformPoseCommandGlobalCfg

Configuration for the command generator.

command property

command: Tensor

The desired 3D-pose in base frame. Shape is (num_envs, 4).

UniformPoseCommandGlobalCfg

Bases: CommandTermCfg

Configuration for the uniform 3D-pose command generator.

Classes:

Name	Description
Ranges	Uniform distribution ranges for the position commands.

Attributes:

Name	Type	Description
asset_name	str	Name of the asset in the environment for which the commands are generated.
simple_heading	bool	Whether to use simple heading or not.
ranges	Ranges	Distribution ranges for the position commands.
goal_pose_visualizer_cfg	VisualizationMarkersCfg	The configuration for the goal pose visualization marker. Defaults to GREEN_ARROW_X_MARKER_CFG.

asset_name class-attribute instance-attribute

asset_name: str = MISSING

Name of the asset in the environment for which the commands are generated.

simple_heading class-attribute instance-attribute

simple_heading: bool = MISSING

Whether to use simple heading or not.

If True, the heading is in the direction of the target position.

ranges class-attribute instance-attribute

ranges: Ranges = MISSING

Distribution ranges for the position commands.

goal_pose_visualizer_cfg class-attribute instance-attribute

goal_pose_visualizer_cfg: VisualizationMarkersCfg = replace(prim_path='/Visuals/Command/pose_goal')

The configuration for the goal pose visualization marker. Defaults to GREEN_ARROW_X_MARKER_CFG.

Ranges

Uniform distribution ranges for the position commands.

Attributes:

Name	Type	Description
pos_x	tuple[float, float]	Range for the x position (in m).
pos_y	tuple[float, float]	Range for the y position (in m).
pos_z	tuple[float, float]	Range for the z position (in m).
heading	tuple[float, float]	Heading range for the position commands (in rad).

pos_x class-attribute instance-attribute

pos_x: tuple[float, float] = MISSING

Range for the x position (in m).

pos_y class-attribute instance-attribute

pos_y: tuple[float, float] = MISSING

Range for the y position (in m).

pos_z class-attribute instance-attribute

pos_z: tuple[float, float] = MISSING

Range for the z position (in m).

heading class-attribute instance-attribute

heading: tuple[float, float] = MISSING

Heading range for the position commands (in rad).

Used only if :attr:simple_heading is False.

UniformVelocityGlobalCommand

UniformVelocityGlobalCommand(cfg: UniformVelocityGlobalCommandCfg, env: ManagerBasedEnv)

Bases: CommandTerm

Command generator that generates a velocity command with uniform distribution.

The command comprises linear velocity in x, y, z directions and an angular velocity around the z-axis. It is given in the robot's base frame.

If the :attr:cfg.heading_command flag is set to True, the angular velocity is computed from the heading error similar to doing a proportional control on the heading error. The target heading is sampled uniformly from the provided range. Otherwise, the angular velocity is sampled uniformly from the provided range.

Mathematically, the angular velocity is computed as follows from the heading command:

.. math::

\omega_z = \frac{1}{2} \text{wrap_to_pi}(\theta_{\text{target}} - \theta_{\text{current}})

Initialize the command generator.

Parameters:

Name	Type	Description	Default
cfg	UniformVelocityGlobalCommandCfg	The configuration of the command generator.	required
env	ManagerBasedEnv	The environment.	required

Raises:

Type	Description
ValueError	If the heading command is active but the heading range is not provided.

Methods:

Name	Description
__str__	Return a string representation of the command generator.

Attributes:

Name	Type	Description
cfg	UniformVelocityGlobalCommandCfg	The configuration of the command generator.
command	Tensor	The desired base velocity command in the base frame. Shape is (num_envs, 4).

cfg instance-attribute

cfg: UniformVelocityGlobalCommandCfg

The configuration of the command generator.

command property

command: Tensor

The desired base velocity command in the base frame. Shape is (num_envs, 4).

__str__

__str__() -> str

Return a string representation of the command generator.

UniformVelocityGlobalCommandCfg

Bases: CommandTermCfg

Configuration for the uniform velocity command generator.

Classes:

Name	Description
Ranges	Uniform distribution ranges for the velocity commands.

Attributes:

Name	Type	Description
asset_name	str	Name of the asset in the environment for which the commands are generated.
heading_command	bool	Whether to use heading command or angular velocity command. Defaults to False.
heading_control_stiffness	float	Scale factor to convert the heading error to angular velocity command. Defaults to 1.0.
rel_standing_envs	float	The sampled probability of environments that should be standing still. Defaults to 0.0.
rel_heading_envs	float	The sampled probability of environments where the robots follow the heading-based angular velocity command
ranges	Ranges	Distribution ranges for the velocity commands.
goal_vel_visualizer_cfg	VisualizationMarkersCfg	The configuration for the goal velocity visualization marker. Defaults to GREEN_ARROW_X_MARKER_CFG.
current_vel_visualizer_cfg	VisualizationMarkersCfg	The configuration for the current velocity visualization marker. Defaults to BLUE_ARROW_X_MARKER_CFG.

asset_name class-attribute instance-attribute

asset_name: str = MISSING

Name of the asset in the environment for which the commands are generated.

heading_command class-attribute instance-attribute

heading_command: bool = False

Whether to use heading command or angular velocity command. Defaults to False.

If True, the angular velocity command is computed from the heading error, where the target heading is sampled uniformly from provided range. Otherwise, the angular velocity command is sampled uniformly from provided range.

heading_control_stiffness class-attribute instance-attribute

heading_control_stiffness: float = 1.0

Scale factor to convert the heading error to angular velocity command. Defaults to 1.0.

rel_standing_envs class-attribute instance-attribute

rel_standing_envs: float = 0.0

The sampled probability of environments that should be standing still. Defaults to 0.0.

rel_heading_envs class-attribute instance-attribute

rel_heading_envs: float = 1.0

The sampled probability of environments where the robots follow the heading-based angular velocity command (the others follow the sampled angular velocity command). Defaults to 1.0.

This parameter is only used if :attr:heading_command is True.

ranges class-attribute instance-attribute

ranges: Ranges = MISSING

Distribution ranges for the velocity commands.

goal_vel_visualizer_cfg class-attribute instance-attribute

goal_vel_visualizer_cfg: VisualizationMarkersCfg = replace(prim_path='/Visuals/Command/velocity_goal')

The configuration for the goal velocity visualization marker. Defaults to GREEN_ARROW_X_MARKER_CFG.

current_vel_visualizer_cfg class-attribute instance-attribute

current_vel_visualizer_cfg: VisualizationMarkersCfg = replace(prim_path='/Visuals/Command/velocity_current')

The configuration for the current velocity visualization marker. Defaults to BLUE_ARROW_X_MARKER_CFG.

Ranges

Uniform distribution ranges for the velocity commands.

Attributes:

Name	Type	Description
lin_vel_x	tuple[float, float]	Range for the linear-x velocity command (in m/s).
lin_vel_y	tuple[float, float]	Range for the linear-y velocity command (in m/s).
lin_vel_z	tuple[float, float]	Range for the linear-z velocity command (in m/s).
ang_vel_z	tuple[float, float]	Range for the angular-z velocity command (in rad/s).
heading	tuple[float, float] | None	Range for the heading command (in rad). Defaults to None.

lin_vel_x class-attribute instance-attribute

lin_vel_x: tuple[float, float] = MISSING

Range for the linear-x velocity command (in m/s).

lin_vel_y class-attribute instance-attribute

lin_vel_y: tuple[float, float] = MISSING

Range for the linear-y velocity command (in m/s).

lin_vel_z class-attribute instance-attribute

lin_vel_z: tuple[float, float] = MISSING

Range for the linear-z velocity command (in m/s).

ang_vel_z class-attribute instance-attribute

ang_vel_z: tuple[float, float] = MISSING

Range for the angular-z velocity command (in rad/s).

heading class-attribute instance-attribute

heading: tuple[float, float] | None = None

Range for the heading command (in rad). Defaults to None.

This parameter is only used if :attr:~UniformVelocityGlobalCommandCfg.heading_command is True.

UniformBimodalVelocityCommand

UniformBimodalVelocityCommand(cfg: UniformBimodalVelocityCommandCfg, env: ManagerBasedEnv)

Bases: CommandTerm

Bimodal velocity command generator with a discrete mode indicator (4+1).

This command samples base-frame velocities with two modes: - Flight mode: uses ranges_flight - Ground mode: uses ranges_ground

The output command augments the 4-D velocity with a 1-D discrete indicator: - 0.0 for flight mode - 1.0 for ground/terrestrial mode

Heading behavior matches UniformVelocityGlobalCommand with optional heading-target proportional control. Ranges are applied per selected mode.

Attributes:

Name	Type	Description
command	Tensor	Return 4-D velocity plus 1-D mode indicator. Shape: (num_envs, 5).

command property

command: Tensor

Return 4-D velocity plus 1-D mode indicator. Shape: (num_envs, 5).

UniformBimodalVelocityCommandCfg

Bases: CommandTermCfg

Configuration for the bimodal velocity command generator (flight vs ground).

Classes:

Name	Description
Ranges	Uniform distribution ranges for velocity commands per mode.

Attributes:

Name	Type	Description
asset_name	str	Name of the asset receiving commands.
heading_command	bool	Whether to convert heading error to angular velocity (per-mode ranges apply).
heading_control_stiffness	float	Scale factor for heading error to yaw rate conversion.
rel_standing_envs	float	Probability that an environment should be standing still (zero command).
rel_heading_envs	float	Probability that an environment uses heading-based angular velocity when enabled.
rel_flight_envs	float	Probability of sampling flight mode (vs ground).
ranges_flight	Ranges	Ranges for flight mode sampling.
ranges_ground	Ranges	Ranges for ground/terrestrial mode sampling.
goal_vel_visualizer_cfg	VisualizationMarkersCfg	The configuration for the goal velocity visualization marker. Defaults to GREEN_ARROW_X_MARKER_CFG.
current_vel_visualizer_cfg	VisualizationMarkersCfg	The configuration for the current velocity visualization marker. Defaults to BLUE_ARROW_X_MARKER_CFG.

asset_name class-attribute instance-attribute

asset_name: str = MISSING

Name of the asset receiving commands.

heading_command class-attribute instance-attribute

heading_command: bool = False

Whether to convert heading error to angular velocity (per-mode ranges apply).

heading_control_stiffness class-attribute instance-attribute

heading_control_stiffness: float = 1.0

Scale factor for heading error to yaw rate conversion.

rel_standing_envs class-attribute instance-attribute

rel_standing_envs: float = 0.0

Probability that an environment should be standing still (zero command).

rel_heading_envs class-attribute instance-attribute

rel_heading_envs: float = 1.0

Probability that an environment uses heading-based angular velocity when enabled.

rel_flight_envs class-attribute instance-attribute

rel_flight_envs: float = 0.5

Probability of sampling flight mode (vs ground).

ranges_flight class-attribute instance-attribute

ranges_flight: Ranges = MISSING

Ranges for flight mode sampling.

ranges_ground class-attribute instance-attribute

ranges_ground: Ranges = MISSING

Ranges for ground/terrestrial mode sampling.

goal_vel_visualizer_cfg class-attribute instance-attribute

goal_vel_visualizer_cfg: VisualizationMarkersCfg = VisualizationMarkersCfg(prim_path='/Visuals/Command/velocity_goal', markers={'flight': UsdFileCfg(usd_path=f'{ISAAC_NUCLEUS_DIR}/Props/UIElements/arrow_x.usd', scale=(1.0, 0.1, 0.1), visual_material=PreviewSurfaceCfg(diffuse_color=(0.0, 1.0, 0.0))), 'ground': UsdFileCfg(usd_path=f'{ISAAC_NUCLEUS_DIR}/Props/UIElements/arrow_x.usd', scale=(1.0, 0.1, 0.1), visual_material=PreviewSurfaceCfg(diffuse_color=(1.0, 0.0, 0.0)))})

The configuration for the goal velocity visualization marker. Defaults to GREEN_ARROW_X_MARKER_CFG.

current_vel_visualizer_cfg class-attribute instance-attribute

current_vel_visualizer_cfg: VisualizationMarkersCfg = replace(prim_path='/Visuals/Command/velocity_current')

The configuration for the current velocity visualization marker. Defaults to BLUE_ARROW_X_MARKER_CFG.

Ranges

Uniform distribution ranges for velocity commands per mode.

Attributes:

Name	Type	Description
lin_vel_x	tuple[float, float]	Range for the linear-x velocity command (in m/s).
lin_vel_y	tuple[float, float]	Range for the linear-y velocity command (in m/s).
lin_vel_z	tuple[float, float]	Range for the linear-z velocity command (in m/s).
ang_vel_z	tuple[float, float]	Range for the angular-z velocity command (in rad/s).
heading	tuple[float, float] | None	Range for the heading command (in rad). Defaults to None.

lin_vel_x class-attribute instance-attribute

lin_vel_x: tuple[float, float] = MISSING

Range for the linear-x velocity command (in m/s).

lin_vel_y class-attribute instance-attribute

lin_vel_y: tuple[float, float] = MISSING

Range for the linear-y velocity command (in m/s).

lin_vel_z class-attribute instance-attribute

lin_vel_z: tuple[float, float] = MISSING

Range for the linear-z velocity command (in m/s).

ang_vel_z class-attribute instance-attribute

ang_vel_z: tuple[float, float] = MISSING

Range for the angular-z velocity command (in rad/s).

heading class-attribute instance-attribute

heading: tuple[float, float] | None = None

Range for the heading command (in rad). Defaults to None.

This parameter is only used if :attr:~UniformBimodalVelocityCommandCfg.heading_command is True.

randomize_action_term_attr

randomize_action_term_attr(env, env_ids: Tensor | None, action_term_name: str = 'control_action', attr_name: str = '_flight_controller', **kwargs)

Randomize a selected action-term attribute.

randomize_controller_params

randomize_controller_params(env, env_ids: Tensor | None, action_term_name: str = 'control_action', controller_attr: str = '_flight_controller', gains: dict | None = None, limits: dict | None = None)

Randomize controller gains and command limits attached to an action term.

randomize_mixer_params

randomize_mixer_params(env, env_ids: Tensor | None, action_term_name: str = 'control_action', mixer_attr: str = '_mixer', param_range: dict[str, tuple[float, float]] | None = None)

Randomize mixer parameters attached to an action term.

randomize_rotor_params

randomize_rotor_params(env, env_ids: Tensor | None, action_term_name: str = 'control_action', rotor_attr: str = '_rotor_dynamics', param_range: dict[str, tuple[float, float]] | None = None)

Randomize rotor parameters attached to an action term.

target_pos_b

target_pos_b(env: ManagerBasedRLEnv, command_name: str | None = None, asset_cfg: SceneEntityCfg = SceneEntityCfg('robot')) -> torch.Tensor

Position of target in body frame.

root_euler_w

root_euler_w(env: ManagerBasedRLEnv, asset_cfg: SceneEntityCfg = SceneEntityCfg('robot')) -> torch.Tensor

Euler angles of the root in world frame.

lin_vel_l2

lin_vel_l2(env: ManagerBasedRLEnv, asset_cfg: SceneEntityCfg = SceneEntityCfg('robot')) -> torch.Tensor

Penalize base linear velocity using L2 squared kernel.

ang_vel_l2

ang_vel_l2(env: ManagerBasedRLEnv, asset_cfg: SceneEntityCfg = SceneEntityCfg('robot')) -> torch.Tensor

Penalize base angular velocity using L2 squared kernel.

pos_error_l2

pos_error_l2(env: ManagerBasedRLEnv, command_name: str) -> torch.Tensor

Penalize asset pos from its target pos using L2 squared kernel.

pos_error_tanh

pos_error_tanh(env: ManagerBasedRLEnv, std: float, command_name: str) -> torch.Tensor

Penalize asset pos from its target pos using tanh kernel.

yaw_error_l2

yaw_error_l2(env: ManagerBasedRLEnv, command_name: str) -> torch.Tensor

Penalize heading error from target heading using L2 squared kernel.

yaw_error_tanh

yaw_error_tanh(env: ManagerBasedRLEnv, std: float, command_name: str) -> torch.Tensor

Penalize heading error from target heading using tanh kernel.

track_lin_vel_z_exp

track_lin_vel_z_exp(env: ManagerBasedRLEnv, std: float, command_name: str, is_bimodal: bool = False, asset_cfg: SceneEntityCfg = SceneEntityCfg('robot')) -> torch.Tensor

Reward tracking of linear velocity commands (z axis) using exponential kernel.

track_lin_vel_exp

track_lin_vel_exp(env: ManagerBasedRLEnv, std: float, command_name: str, asset_cfg: SceneEntityCfg = SceneEntityCfg('robot')) -> torch.Tensor

Reward tracking of linear velocity commands using exponential kernel.

track_yaw_vel_exp

track_yaw_vel_exp(env: ManagerBasedRLEnv, std: float, command_name: str, asset_cfg: SceneEntityCfg = SceneEntityCfg('robot')) -> torch.Tensor

Reward tracking of angular velocity commands (yaw) using exponential kernel.

bimodal_action_tanh

bimodal_action_tanh(env: ManagerBasedRLEnv, std: float, command_name: str, flight_action_name: str = 'control_action', ground_action_name: str = 'track_control_action', ground_weight: float = 1.0, flight_weight: float = 1.0) -> torch.Tensor

Penalize bimodal actions using tanh kernel.

bimodal_height_tanh

bimodal_height_tanh(env: ManagerBasedRLEnv, std: float, command_name: str, asset_cfg: SceneEntityCfg = SceneEntityCfg('robot'), ground_weight: float = 1.0, flight_weight: float = 1.0) -> torch.Tensor

Penalize bimodal height using tanh kernel.

contact_impulse

contact_impulse(env: ManagerBasedRLEnv, threshold: float, sensor_cfg: SceneEntityCfg, mode: str = 'threshold') -> torch.Tensor

Penalize excessive contact impulse (rate of change of contact forces).

This function calculates the impulse as the change in contact forces between consecutive time steps and penalizes values that exceed a threshold.

Parameters:

Name	Type	Description	Default
env	ManagerBasedRLEnv	The learning environment.	required
threshold	float	Maximum acceptable impulse magnitude. Forces below this are not penalized.	required
sensor_cfg	SceneEntityCfg	Configuration for the contact sensor, specifying which bodies to monitor.	required
mode	str	Penalty calculation mode: - "threshold": Penalize only the amount exceeding threshold (continuous) - "binary": Return 1.0 if any impulse exceeds threshold, else 0.0 (discrete) - "total": Return total impulse magnitude regardless of threshold (for monitoring)	'threshold'

Returns:

Type	Description
Tensor	torch.Tensor: Penalty value for each environment: - "threshold" mode: Sum of (impulse - threshold) for all violations - "binary" mode: 1.0 if violation exists, 0.0 otherwise - "total" mode: Total impulse magnitude

Raises:

Type	Description
ValueError	If the sensor history length is less than 2 or the mode is invalid.

Examples:

>>> # Penalize hard landings (impulse > 50 N·s)
>>> impulse_penalty = contact_impulse(
...     env, threshold=50.0,
...     sensor_cfg=SceneEntityCfg("contact_sensor", body_ids=[arml_id, armr_id]),
...     mode="threshold"
... )
>>> rewards["impulse_penalty"] = impulse_penalty * -1.0

bimodal_contacts

bimodal_contacts(env: ManagerBasedRLEnv, command_name: str, threshold: float, sensor_cfg: SceneEntityCfg, mode: str = 'threshold', ground_weight: float = 1.0, flight_weight: float = 1.0) -> torch.Tensor

Penalize contacts when switching from flight to ground mode.

died

died(env: ManagerBasedRLEnv, asset_cfg: SceneEntityCfg = SceneEntityCfg('robot'), min: float = 0.2, max: float = 2.0) -> torch.Tensor

Termination condition based on the robot's height.

actions

Classes:

Name	Description
ControlAction	Body torque control action term.
ControlActionCfg	See :class:ControlAction for more details.
TrackControlAction	Track force control action term.
TrackControlActionCfg	See :class:ControlAction for more details.

ControlAction

ControlAction(cfg: ControlActionCfg, env: ManagerBasedRLEnv)

Bases: ActionTerm

Body torque control action term.

This action term applies a wrench to the drone body frame based on action commands

Attributes:

Name	Type	Description
cfg	ControlActionCfg	The configuration of the action term.

cfg instance-attribute

cfg: ControlActionCfg = cfg

The configuration of the action term.

ControlActionCfg

Bases: ActionTermCfg

See :class:ControlAction for more details.

Attributes:

Name	Type	Description
class_type	type[ActionTerm]	Class of the action term.
asset_name	str	Name of the asset in the environment for which the commands are generated.
params	VehicleParams	Vehicle parameters.
gains	dict | None	Optional gains configuration for the flight controller. See FlightController for details.
limits	dict | None	Optional command-limit configuration passed directly to the flight controller.
control_mode	str	Control mode, options: cmd_motor_thrusts, cmd_motor_speeds, cmd_ctbm, cmd_ctbr, cmd_ctatt, cmd_vel, cmd_pos.

class_type class-attribute instance-attribute

class_type: type[ActionTerm] = ControlAction

Class of the action term.

asset_name class-attribute instance-attribute

asset_name: str = 'robot'

Name of the asset in the environment for which the commands are generated.

params class-attribute instance-attribute

params: VehicleParams = VehicleParams()

Vehicle parameters.

gains class-attribute instance-attribute

gains: dict | None = None

Optional gains configuration for the flight controller. See FlightController for details.

limits class-attribute instance-attribute

limits: dict | None = None

Optional command-limit configuration passed directly to the flight controller.

control_mode class-attribute instance-attribute

control_mode: str = 'cmd_motor_speeds'

Control mode, options: cmd_motor_thrusts, cmd_motor_speeds, cmd_ctbm, cmd_ctbr, cmd_ctatt, cmd_vel, cmd_pos.

rotorpy ref: - https://github.com/spencerfolk/rotorpy/blob/main/rotorpy/learning/quadrotor_environments.py#L319C1-L350C12 - https://github.com/spencerfolk/rotorpy/blob/main/rotorpy/vehicles/multirotor.py#L997

TODO: also align control modes with px4, see: https://docs.px4.io/v1.14/en/flight_modes/offboard (version is unclear ATM)

TrackControlAction

TrackControlAction(cfg: TrackControlActionCfg, env: ManagerBasedRLEnv)

Bases: ActionTerm

Track force control action term.

This action term converts normalized track commands into ground reaction forces applied on each track body.

Attributes:

Name	Type	Description
cfg	TrackControlActionCfg	The configuration of the action term.

cfg instance-attribute

cfg: TrackControlActionCfg = cfg

The configuration of the action term.

TrackControlActionCfg

Bases: ActionTermCfg

See :class:ControlAction for more details.

Attributes:

Name	Type	Description
class_type	type[ActionTerm]	Class of the action term.
asset_name	str	Name of the asset in the environment for which the commands are generated.
params	VehicleParams	Vehicle parameters.

class_type class-attribute instance-attribute

class_type: type[ActionTerm] = TrackControlAction

Class of the action term.

asset_name class-attribute instance-attribute

asset_name: str = 'robot'

Name of the asset in the environment for which the commands are generated.

params class-attribute instance-attribute

params: VehicleParams = VehicleParams()

Vehicle parameters.

commands

Classes:

Name	Description
UniformPoseCommandGlobal	Command generator that generates pose commands containing a 3-D position and heading.
UniformPoseCommandGlobalCfg	Configuration for the uniform 3D-pose command generator.
UniformVelocityGlobalCommand	Command generator that generates a velocity command with uniform distribution.
UniformVelocityGlobalCommandCfg	Configuration for the uniform velocity command generator.
UniformBimodalVelocityCommand	Bimodal velocity command generator with a discrete mode indicator (4+1).
UniformBimodalVelocityCommandCfg	Configuration for the bimodal velocity command generator (flight vs ground).

UniformPoseCommandGlobal

UniformPoseCommandGlobal(cfg: UniformPoseCommandGlobalCfg, env: ManagerBasedEnv)

Bases: CommandTerm

Command generator that generates pose commands containing a 3-D position and heading.

The command generator samples uniform 3D positions around the environment origin. The heading command is either set to point towards the target or is sampled uniformly. This can be configured through the :attr:UniformPoseCommandGlobalCfg.simple_heading parameter in the configuration.

Initialize the command generator class.

Parameters:

Name	Type	Description	Default
cfg	UniformPoseCommandGlobalCfg	The configuration parameters for the command generator.	required
env	ManagerBasedEnv	The environment object.	required

Attributes:

Name	Type	Description
cfg	UniformPoseCommandGlobalCfg	Configuration for the command generator.
command	Tensor	The desired 3D-pose in base frame. Shape is (num_envs, 4).

cfg instance-attribute

cfg: UniformPoseCommandGlobalCfg

Configuration for the command generator.

command property

command: Tensor

The desired 3D-pose in base frame. Shape is (num_envs, 4).

UniformPoseCommandGlobalCfg

Bases: CommandTermCfg

Configuration for the uniform 3D-pose command generator.

Classes:

Name	Description
Ranges	Uniform distribution ranges for the position commands.

Attributes:

Name	Type	Description
asset_name	str	Name of the asset in the environment for which the commands are generated.
simple_heading	bool	Whether to use simple heading or not.
ranges	Ranges	Distribution ranges for the position commands.
goal_pose_visualizer_cfg	VisualizationMarkersCfg	The configuration for the goal pose visualization marker. Defaults to GREEN_ARROW_X_MARKER_CFG.

asset_name class-attribute instance-attribute

asset_name: str = MISSING

Name of the asset in the environment for which the commands are generated.

simple_heading class-attribute instance-attribute

simple_heading: bool = MISSING

Whether to use simple heading or not.

If True, the heading is in the direction of the target position.

ranges class-attribute instance-attribute

ranges: Ranges = MISSING

Distribution ranges for the position commands.

goal_pose_visualizer_cfg class-attribute instance-attribute

goal_pose_visualizer_cfg: VisualizationMarkersCfg = replace(prim_path='/Visuals/Command/pose_goal')

The configuration for the goal pose visualization marker. Defaults to GREEN_ARROW_X_MARKER_CFG.

Ranges

Uniform distribution ranges for the position commands.

Attributes:

Name	Type	Description
pos_x	tuple[float, float]	Range for the x position (in m).
pos_y	tuple[float, float]	Range for the y position (in m).
pos_z	tuple[float, float]	Range for the z position (in m).
heading	tuple[float, float]	Heading range for the position commands (in rad).

pos_x class-attribute instance-attribute

pos_x: tuple[float, float] = MISSING

Range for the x position (in m).

pos_y class-attribute instance-attribute

pos_y: tuple[float, float] = MISSING

Range for the y position (in m).

pos_z class-attribute instance-attribute

pos_z: tuple[float, float] = MISSING

Range for the z position (in m).

heading class-attribute instance-attribute

heading: tuple[float, float] = MISSING

Heading range for the position commands (in rad).

Used only if :attr:simple_heading is False.

UniformVelocityGlobalCommand

UniformVelocityGlobalCommand(cfg: UniformVelocityGlobalCommandCfg, env: ManagerBasedEnv)

Bases: CommandTerm

Command generator that generates a velocity command with uniform distribution.

The command comprises linear velocity in x, y, z directions and an angular velocity around the z-axis. It is given in the robot's base frame.

If the :attr:cfg.heading_command flag is set to True, the angular velocity is computed from the heading error similar to doing a proportional control on the heading error. The target heading is sampled uniformly from the provided range. Otherwise, the angular velocity is sampled uniformly from the provided range.

Mathematically, the angular velocity is computed as follows from the heading command:

.. math::

\omega_z = \frac{1}{2} \text{wrap_to_pi}(\theta_{\text{target}} - \theta_{\text{current}})

Initialize the command generator.

Parameters:

Name	Type	Description	Default
cfg	UniformVelocityGlobalCommandCfg	The configuration of the command generator.	required
env	ManagerBasedEnv	The environment.	required

Raises:

Type	Description
ValueError	If the heading command is active but the heading range is not provided.

Methods:

Name	Description
__str__	Return a string representation of the command generator.

Attributes:

Name	Type	Description
cfg	UniformVelocityGlobalCommandCfg	The configuration of the command generator.
command	Tensor	The desired base velocity command in the base frame. Shape is (num_envs, 4).

cfg instance-attribute

cfg: UniformVelocityGlobalCommandCfg

The configuration of the command generator.

command property

command: Tensor

The desired base velocity command in the base frame. Shape is (num_envs, 4).

__str__

__str__() -> str

Return a string representation of the command generator.

UniformVelocityGlobalCommandCfg

Bases: CommandTermCfg

Configuration for the uniform velocity command generator.

Classes:

Name	Description
Ranges	Uniform distribution ranges for the velocity commands.

Attributes:

Name	Type	Description
asset_name	str	Name of the asset in the environment for which the commands are generated.
heading_command	bool	Whether to use heading command or angular velocity command. Defaults to False.
heading_control_stiffness	float	Scale factor to convert the heading error to angular velocity command. Defaults to 1.0.
rel_standing_envs	float	The sampled probability of environments that should be standing still. Defaults to 0.0.
rel_heading_envs	float	The sampled probability of environments where the robots follow the heading-based angular velocity command
ranges	Ranges	Distribution ranges for the velocity commands.
goal_vel_visualizer_cfg	VisualizationMarkersCfg	The configuration for the goal velocity visualization marker. Defaults to GREEN_ARROW_X_MARKER_CFG.
current_vel_visualizer_cfg	VisualizationMarkersCfg	The configuration for the current velocity visualization marker. Defaults to BLUE_ARROW_X_MARKER_CFG.

asset_name class-attribute instance-attribute

asset_name: str = MISSING

Name of the asset in the environment for which the commands are generated.

heading_command class-attribute instance-attribute

heading_command: bool = False

Whether to use heading command or angular velocity command. Defaults to False.

If True, the angular velocity command is computed from the heading error, where the target heading is sampled uniformly from provided range. Otherwise, the angular velocity command is sampled uniformly from provided range.

heading_control_stiffness class-attribute instance-attribute

heading_control_stiffness: float = 1.0

Scale factor to convert the heading error to angular velocity command. Defaults to 1.0.

rel_standing_envs class-attribute instance-attribute

rel_standing_envs: float = 0.0

The sampled probability of environments that should be standing still. Defaults to 0.0.

rel_heading_envs class-attribute instance-attribute

rel_heading_envs: float = 1.0

The sampled probability of environments where the robots follow the heading-based angular velocity command (the others follow the sampled angular velocity command). Defaults to 1.0.

This parameter is only used if :attr:heading_command is True.

ranges class-attribute instance-attribute

ranges: Ranges = MISSING

Distribution ranges for the velocity commands.

goal_vel_visualizer_cfg class-attribute instance-attribute

goal_vel_visualizer_cfg: VisualizationMarkersCfg = replace(prim_path='/Visuals/Command/velocity_goal')

The configuration for the goal velocity visualization marker. Defaults to GREEN_ARROW_X_MARKER_CFG.

current_vel_visualizer_cfg class-attribute instance-attribute

current_vel_visualizer_cfg: VisualizationMarkersCfg = replace(prim_path='/Visuals/Command/velocity_current')

The configuration for the current velocity visualization marker. Defaults to BLUE_ARROW_X_MARKER_CFG.

Ranges

Uniform distribution ranges for the velocity commands.

Attributes:

Name	Type	Description
lin_vel_x	tuple[float, float]	Range for the linear-x velocity command (in m/s).
lin_vel_y	tuple[float, float]	Range for the linear-y velocity command (in m/s).
lin_vel_z	tuple[float, float]	Range for the linear-z velocity command (in m/s).
ang_vel_z	tuple[float, float]	Range for the angular-z velocity command (in rad/s).
heading	tuple[float, float] | None	Range for the heading command (in rad). Defaults to None.

lin_vel_x class-attribute instance-attribute

lin_vel_x: tuple[float, float] = MISSING

Range for the linear-x velocity command (in m/s).

lin_vel_y class-attribute instance-attribute

lin_vel_y: tuple[float, float] = MISSING

Range for the linear-y velocity command (in m/s).

lin_vel_z class-attribute instance-attribute

lin_vel_z: tuple[float, float] = MISSING

Range for the linear-z velocity command (in m/s).

ang_vel_z class-attribute instance-attribute

ang_vel_z: tuple[float, float] = MISSING

Range for the angular-z velocity command (in rad/s).

heading class-attribute instance-attribute

heading: tuple[float, float] | None = None

Range for the heading command (in rad). Defaults to None.

This parameter is only used if :attr:~UniformVelocityGlobalCommandCfg.heading_command is True.

UniformBimodalVelocityCommand

UniformBimodalVelocityCommand(cfg: UniformBimodalVelocityCommandCfg, env: ManagerBasedEnv)

Bases: CommandTerm

Bimodal velocity command generator with a discrete mode indicator (4+1).

This command samples base-frame velocities with two modes: - Flight mode: uses ranges_flight - Ground mode: uses ranges_ground

The output command augments the 4-D velocity with a 1-D discrete indicator: - 0.0 for flight mode - 1.0 for ground/terrestrial mode

Heading behavior matches UniformVelocityGlobalCommand with optional heading-target proportional control. Ranges are applied per selected mode.

Attributes:

Name	Type	Description
command	Tensor	Return 4-D velocity plus 1-D mode indicator. Shape: (num_envs, 5).

command property

command: Tensor

Return 4-D velocity plus 1-D mode indicator. Shape: (num_envs, 5).

UniformBimodalVelocityCommandCfg

Bases: CommandTermCfg

Configuration for the bimodal velocity command generator (flight vs ground).

Classes:

Name	Description
Ranges	Uniform distribution ranges for velocity commands per mode.

Attributes:

Name	Type	Description
asset_name	str	Name of the asset receiving commands.
heading_command	bool	Whether to convert heading error to angular velocity (per-mode ranges apply).
heading_control_stiffness	float	Scale factor for heading error to yaw rate conversion.
rel_standing_envs	float	Probability that an environment should be standing still (zero command).
rel_heading_envs	float	Probability that an environment uses heading-based angular velocity when enabled.
rel_flight_envs	float	Probability of sampling flight mode (vs ground).
ranges_flight	Ranges	Ranges for flight mode sampling.
ranges_ground	Ranges	Ranges for ground/terrestrial mode sampling.
goal_vel_visualizer_cfg	VisualizationMarkersCfg	The configuration for the goal velocity visualization marker. Defaults to GREEN_ARROW_X_MARKER_CFG.
current_vel_visualizer_cfg	VisualizationMarkersCfg	The configuration for the current velocity visualization marker. Defaults to BLUE_ARROW_X_MARKER_CFG.

asset_name class-attribute instance-attribute

asset_name: str = MISSING

Name of the asset receiving commands.

heading_command class-attribute instance-attribute

heading_command: bool = False

Whether to convert heading error to angular velocity (per-mode ranges apply).

heading_control_stiffness class-attribute instance-attribute

heading_control_stiffness: float = 1.0

Scale factor for heading error to yaw rate conversion.

rel_standing_envs class-attribute instance-attribute

rel_standing_envs: float = 0.0

Probability that an environment should be standing still (zero command).

rel_heading_envs class-attribute instance-attribute

rel_heading_envs: float = 1.0

Probability that an environment uses heading-based angular velocity when enabled.

rel_flight_envs class-attribute instance-attribute

rel_flight_envs: float = 0.5

Probability of sampling flight mode (vs ground).

ranges_flight class-attribute instance-attribute

ranges_flight: Ranges = MISSING

Ranges for flight mode sampling.

ranges_ground class-attribute instance-attribute

ranges_ground: Ranges = MISSING

Ranges for ground/terrestrial mode sampling.

goal_vel_visualizer_cfg class-attribute instance-attribute

goal_vel_visualizer_cfg: VisualizationMarkersCfg = VisualizationMarkersCfg(prim_path='/Visuals/Command/velocity_goal', markers={'flight': UsdFileCfg(usd_path=f'{ISAAC_NUCLEUS_DIR}/Props/UIElements/arrow_x.usd', scale=(1.0, 0.1, 0.1), visual_material=PreviewSurfaceCfg(diffuse_color=(0.0, 1.0, 0.0))), 'ground': UsdFileCfg(usd_path=f'{ISAAC_NUCLEUS_DIR}/Props/UIElements/arrow_x.usd', scale=(1.0, 0.1, 0.1), visual_material=PreviewSurfaceCfg(diffuse_color=(1.0, 0.0, 0.0)))})

The configuration for the goal velocity visualization marker. Defaults to GREEN_ARROW_X_MARKER_CFG.

current_vel_visualizer_cfg class-attribute instance-attribute

current_vel_visualizer_cfg: VisualizationMarkersCfg = replace(prim_path='/Visuals/Command/velocity_current')

The configuration for the current velocity visualization marker. Defaults to BLUE_ARROW_X_MARKER_CFG.

Ranges

Uniform distribution ranges for velocity commands per mode.

Attributes:

Name	Type	Description
lin_vel_x	tuple[float, float]	Range for the linear-x velocity command (in m/s).
lin_vel_y	tuple[float, float]	Range for the linear-y velocity command (in m/s).
lin_vel_z	tuple[float, float]	Range for the linear-z velocity command (in m/s).
ang_vel_z	tuple[float, float]	Range for the angular-z velocity command (in rad/s).
heading	tuple[float, float] | None	Range for the heading command (in rad). Defaults to None.

lin_vel_x class-attribute instance-attribute

lin_vel_x: tuple[float, float] = MISSING

Range for the linear-x velocity command (in m/s).

lin_vel_y class-attribute instance-attribute

lin_vel_y: tuple[float, float] = MISSING

Range for the linear-y velocity command (in m/s).

lin_vel_z class-attribute instance-attribute

lin_vel_z: tuple[float, float] = MISSING

Range for the linear-z velocity command (in m/s).

ang_vel_z class-attribute instance-attribute

ang_vel_z: tuple[float, float] = MISSING

Range for the angular-z velocity command (in rad/s).

heading class-attribute instance-attribute

heading: tuple[float, float] | None = None

Range for the heading command (in rad). Defaults to None.

This parameter is only used if :attr:~UniformBimodalVelocityCommandCfg.heading_command is True.

events

Functions:

Name	Description
randomize_action_term_attr	Randomize a selected action-term attribute.
randomize_controller_params	Randomize controller gains and command limits attached to an action term.
randomize_mixer_params	Randomize mixer parameters attached to an action term.
randomize_rotor_params	Randomize rotor parameters attached to an action term.

randomize_action_term_attr

randomize_action_term_attr(env, env_ids: Tensor | None, action_term_name: str = 'control_action', attr_name: str = '_flight_controller', **kwargs)

Randomize a selected action-term attribute.

randomize_controller_params

randomize_controller_params(env, env_ids: Tensor | None, action_term_name: str = 'control_action', controller_attr: str = '_flight_controller', gains: dict | None = None, limits: dict | None = None)

Randomize controller gains and command limits attached to an action term.

randomize_mixer_params

randomize_mixer_params(env, env_ids: Tensor | None, action_term_name: str = 'control_action', mixer_attr: str = '_mixer', param_range: dict[str, tuple[float, float]] | None = None)

Randomize mixer parameters attached to an action term.

randomize_rotor_params

randomize_rotor_params(env, env_ids: Tensor | None, action_term_name: str = 'control_action', rotor_attr: str = '_rotor_dynamics', param_range: dict[str, tuple[float, float]] | None = None)

Randomize rotor parameters attached to an action term.

observations

Functions:

Name	Description
target_pos_b	Position of target in body frame.
root_euler_w	Euler angles of the root in world frame.

target_pos_b

target_pos_b(env: ManagerBasedRLEnv, command_name: str | None = None, asset_cfg: SceneEntityCfg = SceneEntityCfg('robot')) -> torch.Tensor

Position of target in body frame.

root_euler_w

root_euler_w(env: ManagerBasedRLEnv, asset_cfg: SceneEntityCfg = SceneEntityCfg('robot')) -> torch.Tensor

Euler angles of the root in world frame.

rewards

Functions:

Name	Description
lin_vel_l2	Penalize base linear velocity using L2 squared kernel.
ang_vel_l2	Penalize base angular velocity using L2 squared kernel.
pos_error_l2	Penalize asset pos from its target pos using L2 squared kernel.
pos_error_tanh	Penalize asset pos from its target pos using tanh kernel.
yaw_error_l2	Penalize heading error from target heading using L2 squared kernel.
yaw_error_tanh	Penalize heading error from target heading using tanh kernel.
track_lin_vel_z_exp	Reward tracking of linear velocity commands (z axis) using exponential kernel.
track_lin_vel_exp	Reward tracking of linear velocity commands using exponential kernel.
track_yaw_vel_exp	Reward tracking of angular velocity commands (yaw) using exponential kernel.
bimodal_action_tanh	Penalize bimodal actions using tanh kernel.
bimodal_height_tanh	Penalize bimodal height using tanh kernel.
contact_impulse	Penalize excessive contact impulse (rate of change of contact forces).
bimodal_contacts	Penalize contacts when switching from flight to ground mode.

lin_vel_l2

lin_vel_l2(env: ManagerBasedRLEnv, asset_cfg: SceneEntityCfg = SceneEntityCfg('robot')) -> torch.Tensor

Penalize base linear velocity using L2 squared kernel.

ang_vel_l2

ang_vel_l2(env: ManagerBasedRLEnv, asset_cfg: SceneEntityCfg = SceneEntityCfg('robot')) -> torch.Tensor

Penalize base angular velocity using L2 squared kernel.

pos_error_l2

pos_error_l2(env: ManagerBasedRLEnv, command_name: str) -> torch.Tensor

Penalize asset pos from its target pos using L2 squared kernel.

pos_error_tanh

pos_error_tanh(env: ManagerBasedRLEnv, std: float, command_name: str) -> torch.Tensor

Penalize asset pos from its target pos using tanh kernel.

yaw_error_l2

yaw_error_l2(env: ManagerBasedRLEnv, command_name: str) -> torch.Tensor

Penalize heading error from target heading using L2 squared kernel.

yaw_error_tanh

yaw_error_tanh(env: ManagerBasedRLEnv, std: float, command_name: str) -> torch.Tensor

Penalize heading error from target heading using tanh kernel.

track_lin_vel_z_exp

track_lin_vel_z_exp(env: ManagerBasedRLEnv, std: float, command_name: str, is_bimodal: bool = False, asset_cfg: SceneEntityCfg = SceneEntityCfg('robot')) -> torch.Tensor

Reward tracking of linear velocity commands (z axis) using exponential kernel.

track_lin_vel_exp

track_lin_vel_exp(env: ManagerBasedRLEnv, std: float, command_name: str, asset_cfg: SceneEntityCfg = SceneEntityCfg('robot')) -> torch.Tensor

Reward tracking of linear velocity commands using exponential kernel.

track_yaw_vel_exp

track_yaw_vel_exp(env: ManagerBasedRLEnv, std: float, command_name: str, asset_cfg: SceneEntityCfg = SceneEntityCfg('robot')) -> torch.Tensor

Reward tracking of angular velocity commands (yaw) using exponential kernel.

bimodal_action_tanh

bimodal_action_tanh(env: ManagerBasedRLEnv, std: float, command_name: str, flight_action_name: str = 'control_action', ground_action_name: str = 'track_control_action', ground_weight: float = 1.0, flight_weight: float = 1.0) -> torch.Tensor

Penalize bimodal actions using tanh kernel.

bimodal_height_tanh

bimodal_height_tanh(env: ManagerBasedRLEnv, std: float, command_name: str, asset_cfg: SceneEntityCfg = SceneEntityCfg('robot'), ground_weight: float = 1.0, flight_weight: float = 1.0) -> torch.Tensor

Penalize bimodal height using tanh kernel.

contact_impulse

contact_impulse(env: ManagerBasedRLEnv, threshold: float, sensor_cfg: SceneEntityCfg, mode: str = 'threshold') -> torch.Tensor

Penalize excessive contact impulse (rate of change of contact forces).

This function calculates the impulse as the change in contact forces between consecutive time steps and penalizes values that exceed a threshold.

Parameters:

Name	Type	Description	Default
env	ManagerBasedRLEnv	The learning environment.	required
threshold	float	Maximum acceptable impulse magnitude. Forces below this are not penalized.	required
sensor_cfg	SceneEntityCfg	Configuration for the contact sensor, specifying which bodies to monitor.	required
mode	str	Penalty calculation mode: - "threshold": Penalize only the amount exceeding threshold (continuous) - "binary": Return 1.0 if any impulse exceeds threshold, else 0.0 (discrete) - "total": Return total impulse magnitude regardless of threshold (for monitoring)	'threshold'

Returns:

Type	Description
Tensor	torch.Tensor: Penalty value for each environment: - "threshold" mode: Sum of (impulse - threshold) for all violations - "binary" mode: 1.0 if violation exists, 0.0 otherwise - "total" mode: Total impulse magnitude

Raises:

Type	Description
ValueError	If the sensor history length is less than 2 or the mode is invalid.

Examples:

>>> # Penalize hard landings (impulse > 50 N·s)
>>> impulse_penalty = contact_impulse(
...     env, threshold=50.0,
...     sensor_cfg=SceneEntityCfg("contact_sensor", body_ids=[arml_id, armr_id]),
...     mode="threshold"
... )
>>> rewards["impulse_penalty"] = impulse_penalty * -1.0

bimodal_contacts

bimodal_contacts(env: ManagerBasedRLEnv, command_name: str, threshold: float, sensor_cfg: SceneEntityCfg, mode: str = 'threshold', ground_weight: float = 1.0, flight_weight: float = 1.0) -> torch.Tensor

Penalize contacts when switching from flight to ground mode.

terminations

Functions:

Name	Description
died	Termination condition based on the robot's height.

died

died(env: ManagerBasedRLEnv, asset_cfg: SceneEntityCfg = SceneEntityCfg('robot'), min: float = 0.2, max: float = 2.0) -> torch.Tensor

Termination condition based on the robot's height.

LAV2_base_direct

Modules:

Name	Description
quadcopter_env

quadcopter_env

Classes:

Name	Description
QuadcopterEnvWindow	Window manager for the Quadcopter environment.

QuadcopterEnvWindow

QuadcopterEnvWindow(env: QuadcopterEnv, window_name: str = 'IsaacLab')

Bases: BaseEnvWindow

Window manager for the Quadcopter environment.

Initialize the window.

Parameters:

Name	Type	Description	Default
env	QuadcopterEnv	The environment object.	required
window_name	str	The name of the window. Defaults to "IsaacLab".	'IsaacLab'

LAV2_navrl

Modules:

Name	Description
LAV2_navrl_env

LAV2_navrl_env

Classes:

Name	Description
LAV2NavRLEnvWindow	Window manager for the Quadcopter environment.

LAV2NavRLEnvWindow

LAV2NavRLEnvWindow(env: LAV2NavRLEnv, window_name: str = 'IsaacLab')

Bases: BaseEnvWindow

Window manager for the Quadcopter environment.

Initialize the window.

Parameters:

Name	Type	Description	Default
env	LAV2NavRLEnv	The environment object.	required
window_name	str	The name of the window. Defaults to "IsaacLab".	'IsaacLab'

LAV2_trajectory

Modules:

Name	Description
LAV2_traj_env_cfg
mdp

LAV2_traj_env_cfg

Classes:

Name	Description
ActionsCfg	Action specifications for the MDP.
EventCfg	Configuration for events.
TerminationsCfg	Termination terms for the MDP.

ActionsCfg

Action specifications for the MDP.

EventCfg

Configuration for events.

TerminationsCfg

Termination terms for the MDP.

mdp

Modules:

Name	Description
commands

Classes:

Name	Description
FlightActionMapper	Map batched normalized actions into controller targets, wrench, or RPM.
FlightMappingLimits	Per-channel limits used when mapping normalized actions to targets.
Mixer	Map thrust and moment demands to rotor commands.
FlightController	PID flight controller (batched).
VehicleParams	LAV2 vehicle parameters.
RotorDynamics	Batched rotor dynamics model.
TrackDynamics	Batched track dynamics model.
ControlAction	Body torque control action term.
ControlActionCfg	See :class:ControlAction for more details.
TrackControlAction	Track force control action term.
TrackControlActionCfg	See :class:ControlAction for more details.
UniformPoseCommandGlobal	Command generator that generates pose commands containing a 3-D position and heading.
UniformPoseCommandGlobalCfg	Configuration for the uniform 3D-pose command generator.
UniformVelocityGlobalCommand	Command generator that generates a velocity command with uniform distribution.
UniformVelocityGlobalCommandCfg	Configuration for the uniform velocity command generator.
UniformBimodalVelocityCommand	Bimodal velocity command generator with a discrete mode indicator (4+1).
UniformBimodalVelocityCommandCfg	Configuration for the bimodal velocity command generator (flight vs ground).
HelixTrajectory	Batched helix trajectory with optional vertical oscillation.
LemniscateTrajectory	Batched Bernoulli lemniscate trajectory with vertical oscillation.
LissajousTrajectory	Batched three-dimensional Lissajous trajectory.
RectangleTrajectory	Batched constant-speed rectangle trajectory in the horizontal plane.
RandomTrajectoryCommand	Trajectory command generator with randomized type and parameters per environment.
RandomTrajectoryCommandCfg	Configuration for randomized trajectory commands.

Functions:

Name	Description
randomize_action_term_attr	Randomize a selected action-term attribute.
randomize_controller_params	Randomize controller gains and command limits attached to an action term.
randomize_mixer_params	Randomize mixer parameters attached to an action term.
randomize_rotor_params	Randomize rotor parameters attached to an action term.
target_pos_b	Position of target in body frame.
root_euler_w	Euler angles of the root in world frame.
lin_vel_l2	Penalize base linear velocity using L2 squared kernel.
ang_vel_l2	Penalize base angular velocity using L2 squared kernel.
pos_error_l2	Penalize asset pos from its target pos using L2 squared kernel.
pos_error_tanh	Penalize asset pos from its target pos using tanh kernel.
yaw_error_l2	Penalize heading error from target heading using L2 squared kernel.
yaw_error_tanh	Penalize heading error from target heading using tanh kernel.
track_lin_vel_z_exp	Reward tracking of linear velocity commands (z axis) using exponential kernel.
track_lin_vel_exp	Reward tracking of linear velocity commands using exponential kernel.
track_yaw_vel_exp	Reward tracking of angular velocity commands (yaw) using exponential kernel.
bimodal_action_tanh	Penalize bimodal actions using tanh kernel.
bimodal_height_tanh	Penalize bimodal height using tanh kernel.
contact_impulse	Penalize excessive contact impulse (rate of change of contact forces).
bimodal_contacts	Penalize contacts when switching from flight to ground mode.
died	Termination condition based on the robot's height.

FlightActionMapper

FlightActionMapper(control_mode: str, params: VehicleParams = VehicleParams(), limits: FlightMappingLimits = FlightMappingLimits(), *, num_envs: int = 1, device: str | device = 'cpu')

Map batched normalized actions into controller targets, wrench, or RPM.

Initialize a batched Torch flight action mapper for the selected mode.

Methods:

Name	Description
map_action	Map a normalized batched action into RPM commands and intermediates.

Attributes:

Name	Type	Description
requires_controller	bool	Whether this mode needs a controller target expansion step.

requires_controller property

requires_controller: bool

Whether this mode needs a controller target expansion step.

map_action

map_action(action: Tensor, mixer: Mixer, *, state: Tensor | None = None, flight_controller: ControllerBase | None = None) -> FlightMappingResult

Map a normalized batched action into RPM commands and intermediates.

FlightMappingLimits dataclass

FlightMappingLimits(vel_limits: tuple[float, float, float] = (5.0, 5.0, 5.0), att_limits: tuple[float, float, float] = (1.0, 1.0, math.pi), br_limits: tuple[float, float, float] = (2.0, 2.0, 2.0))

Per-channel limits used when mapping normalized actions to targets.

Mixer

Mixer(params: VehicleParams = VehicleParams(), num_envs: int = 1, device: str | device = 'cpu')

Map thrust and moment demands to rotor commands.

Note

电机推力分配： 电机1 = 推力 - 横滚(roll) + 俯仰(pitch) + 偏航(yaw) 电机2 = 推力 + 横滚(roll) - 俯仰(pitch) + 偏航(yaw) 电机3 = 推力 + 横滚(roll) + 俯仰(pitch) - 偏航(yaw) 电机4 = 推力 - 横滚(roll) - 俯仰(pitch) - 偏航(yaw)

Initialize the torch mixer with vehicle parameters and device.

Methods:

Name	Description
randomize	Randomize mixer runtime parameters from nominal values.
apply_thrust_curve	Apply thrust curve to normalized thrust command.
calculate_rotor_commands	Allocate thrust and moments into rotor speeds.

randomize

randomize(env_ids=None, randomization=None)

Randomize mixer runtime parameters from nominal values.

apply_thrust_curve

apply_thrust_curve(normalized_thrust: Tensor, mode: int = 1) -> torch.Tensor

Apply thrust curve to normalized thrust command.

References

cmd_ctatt: https://github.com/PX4/PX4-Autopilot/blob/e5a483e4b71ab4b994cb0121fe28992d8ce0c8cd/src/modules/mc_att_control/mc_att_control_main.cpp#L111-L137 cmd_ctbr: https://github.com/PX4/PX4-Autopilot/blob/e5a483e4b71ab4b994cb0121fe28992d8ce0c8cd/src/modules/mc_rate_control/MulticopterRateControl.cpp#L156-L179

Parameters:

Name	Type	Description	Default
normalized_thrust	Tensor	Normalized thrust command [-1, 1]. Shape: arbitrary, e.g. (num_envs, n_motors). Applied elementwise.	required
mode	int	Thrust curve mode.	1

Returns:

Type	Description
Tensor	torch.Tensor: Thrust in Newtons with the same shape as normalized_thrust.

Raises:

Type	Description
NotImplementedError	If the specified mode is not implemented.

calculate_rotor_commands

calculate_rotor_commands(control: Tensor) -> torch.Tensor

Allocate thrust and moments into rotor speeds.

Parameters:

Name	Type	Description	Default
control	Tensor	Desired total thrust and roll/pitch/yaw moments. Shape: (num_envs, 4)	required

Returns:

Type	Description
Tensor	torch.Tensor: The amplitude of each motor's rotational speed (calculated using sqrt(clip(rpm_sq))). Shape: (num_envs, 4)

FlightController

FlightController(params: VehicleParams = VehicleParams(), control_mask: list[int] = [1, 1, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0], gains: dict | None = None, limits: dict | None = None, num_envs: int = 1, device: str | device = 'cpu')

Bases: ControllerBase

PID flight controller (batched).

Initialize the batched flight PID controller.

Methods:

Name	Description
randomize	Randomize configured PID gains and command limits for selected environments.
reset	Reset PID loops for selected environments.
update	Update controller.

Attributes:

Name	Type	Description
params	VehicleParams	Vehicle parameters for the controller.
control_mask	ndarray | Tensor	unified control mask for target/state channels, used in update(). e.g. using 9~11 for attitude yaw control.

params instance-attribute

params: VehicleParams = params

Vehicle parameters for the controller.

control_mask instance-attribute

control_mask: ndarray | Tensor = control_mask

unified control mask for target/state channels, used in update(). e.g. using 9~11 for attitude yaw control.

randomize

randomize(env_ids=None, gains=None, limits=None)

Randomize configured PID gains and command limits for selected environments.

reset

reset(env_ids)

Reset PID loops for selected environments.

update

update(target: Tensor, state: Tensor) -> torch.Tensor

Update controller.

Parameters:

Name	Type	Description	Default
target	Tensor	(N, 12) desired [x, y, z, vx, vy, vz, roll, pitch, yaw, wx, wy, wz]	required
state	Tensor	(N, 12) current state same ordering.	required

Returns:

Type	Description
Tensor	torch.Tensor: Tensor of shape (N, 4): [thrust, roll_moment, pitch_moment, yaw_moment].

VehicleParams dataclass

VehicleParams(rho: float = 1.184, g: float = 9.81, sim_dt: float = 0.01, step_dt: float = 0.01, randomization: dict[str, dict[str, tuple[float, float]]] = dict(), mass: float = 2.1, inertia: list = (lambda: np.diag([0.11, 0.108, 0.215]).tolist())(), diameter: float = 0.476, Cdx: float = 0.5, x: float = 0.0952, y: float = 0.114423, h: float = -0.0125, alpha: float = 8 * (np.pi / 180), r_p: float = 0.0775, J_m: float = 1e-05, Ct: float = 0.666, Cq: float = 0.0716, theta0: float = 14.6 * (np.pi / 180), thetaTip: float = 6.8 * (np.pi / 180), lock: float = 0.6051, max_rpm: float = 10000.0, init_rpm: float = -1.0, rotor_rpm_rate_limit: float = 50000.0, tau_f: float = 0.01, tau_m: float = 0.05, tau_up: float = -1.0, tau_down: float = -1.0, r_w: float = 0.0285, n_w: int = 4, B: float = 0.136, L: float = 0.23, h_cg: float = 0.078, f: float = 0.01, mu0: float = 0.8, K: float = 20, track_cmd_rate_limit: float = 100, min_throttle_manual: float = 0.08, min_throttle: float = 0.12, max_throttle: float = 1.0, hover_throttle: float = -1)

LAV2 vehicle parameters.

Methods:

Name	Description
__post_init__	Calculate derived parameters.

Attributes:

Name	Type	Description
rho	float	空气密度 (kg/m^3)
g	float	重力加速度 (m/s^2)
sim_dt	float	仿真步长 (s)
step_dt	float	控制步长 (s)
randomization	dict[str, dict[str, tuple[float, float]]]	运行时参数随机化配置，按组件 (e.g. rotor/mixer) 划分为多个 scale 相对范围子表
mass	float	质量 (kg)
inertia	list	惯性矩阵 (kg*m^2)
diameter	float	直径 (m)
Cdx	float	空气阻力系数
x	float	x方向臂长 (m)
y	float	y方向臂长 (m)
h	float	重心高度 (m)
alpha	float	y方向涵道外倾角度 (rad)
r_p	float	螺旋桨半径 (m)
J_m	float	电机转动惯量 (kg*m^2)
Ct	float	推力系数
Cq	float	扭矩系数
theta0	float	桨根角 (rad)
thetaTip	float	桨尖角 (rad)
theta1	float	桨叶扭转角 (rad)
lock	float	Lock number
max_rpm	float	最大转速
init_rpm	float	初始转速
rotor_rpm_rate_limit	float	旋翼转速变化率限制 (rpm/s)
cT	float	转速平方到推力增益 (N/(rpm^2))
cM	float	转速平方到扭矩增益 (N*m/(rpm^2))
tau_f	float	电机滤波时间常数 (s)
tau_m	float	兼容保留的电机时间常数 (s)
tau_up	float	电机升速时间常数 (s)
tau_down	float	电机降速时间常数 (s)
r_w	float	主动轮半径 (m)
n_w	int	负重轮个数
B	float	两侧履带中心距 (m)
L	float	履带节点接地部分长度 (m)
h_cg	float	车体质心高度 (m)
f	float	滚动阻力系数(以干燥混凝土路面或沥青路面为例)
mu0	float	滑转率为1时的摩擦系数
K	float	依赖于土壤黏聚系数和摩擦特性的常数
track_cmd_rate_limit	float	履带主动轮指令变化率限制 (rad/s^2)
min_throttle_manual	float	MPC_MANTHR_MIN: Minimum manual thrust (0.0 - 1.0). Default: 0.08
min_throttle	float	MPC_THR_MIN: Minimum thrust in auto thrust control (0.05 - 1.0). Default: 0.12
max_throttle	float	MPC_THR_MAX: Maximum thrust in auto thrust control (0.0 - 1.0). Default: 1.0
hover_throttle	float	MPC_THR_HOVER: Hover throttle manual (0.1 - 0.8). Default: 0.5 in PX4, calculated hover throttle in LAV2

rho class-attribute instance-attribute

rho: float = 1.184

空气密度 (kg/m^3)

g class-attribute instance-attribute

g: float = 9.81

重力加速度 (m/s^2)

sim_dt class-attribute instance-attribute

sim_dt: float = 0.01

仿真步长 (s)

step_dt class-attribute instance-attribute

step_dt: float = 0.01

控制步长 (s)

randomization class-attribute instance-attribute

randomization: dict[str, dict[str, tuple[float, float]]] = field(default_factory=dict)

运行时参数随机化配置，按组件 (e.g. rotor/mixer) 划分为多个 scale 相对范围子表

mass class-attribute instance-attribute

mass: float = 2.1

质量 (kg)

inertia class-attribute instance-attribute

inertia: list = field(default_factory=lambda: tolist())

惯性矩阵 (kg*m^2)

diameter class-attribute instance-attribute

diameter: float = 0.476

直径 (m)

Cdx class-attribute instance-attribute

Cdx: float = 0.5

空气阻力系数

x class-attribute instance-attribute

x: float = 0.0952

x方向臂长 (m)

y class-attribute instance-attribute

y: float = 0.114423

y方向臂长 (m)

h class-attribute instance-attribute

h: float = -0.0125

重心高度 (m)

alpha class-attribute instance-attribute

alpha: float = 8 * (pi / 180)

y方向涵道外倾角度 (rad)

r_p class-attribute instance-attribute

r_p: float = 0.0775

螺旋桨半径 (m)

J_m class-attribute instance-attribute

J_m: float = 1e-05

电机转动惯量 (kg*m^2)

Ct class-attribute instance-attribute

Ct: float = 0.666

推力系数

Cq class-attribute instance-attribute

Cq: float = 0.0716

扭矩系数

theta0 class-attribute instance-attribute

theta0: float = 14.6 * (pi / 180)

桨根角 (rad)

thetaTip class-attribute instance-attribute

thetaTip: float = 6.8 * (pi / 180)

桨尖角 (rad)

theta1 class-attribute instance-attribute

theta1: float = field(init=False)

桨叶扭转角 (rad)

lock class-attribute instance-attribute

lock: float = 0.6051

Lock number

max_rpm class-attribute instance-attribute

max_rpm: float = 10000.0

最大转速

init_rpm class-attribute instance-attribute

init_rpm: float = -1.0

初始转速

rotor_rpm_rate_limit class-attribute instance-attribute

rotor_rpm_rate_limit: float = 50000.0

旋翼转速变化率限制 (rpm/s)

cT class-attribute instance-attribute

cT: float = field(init=False)

转速平方到推力增益 (N/(rpm^2))

cM class-attribute instance-attribute

cM: float = field(init=False)

转速平方到扭矩增益 (N*m/(rpm^2))

tau_f class-attribute instance-attribute

tau_f: float = 0.01

电机滤波时间常数 (s)

tau_m class-attribute instance-attribute

tau_m: float = 0.05

兼容保留的电机时间常数 (s)

tau_up class-attribute instance-attribute

tau_up: float = -1.0

电机升速时间常数 (s)

tau_down class-attribute instance-attribute

tau_down: float = -1.0

电机降速时间常数 (s)

r_w class-attribute instance-attribute

r_w: float = 0.0285

主动轮半径 (m)

n_w class-attribute instance-attribute

n_w: int = 4

负重轮个数

B class-attribute instance-attribute

B: float = 0.136

两侧履带中心距 (m)

L class-attribute instance-attribute

L: float = 0.23

履带节点接地部分长度 (m)

h_cg class-attribute instance-attribute

h_cg: float = 0.078

车体质心高度 (m)

f class-attribute instance-attribute

f: float = 0.01

滚动阻力系数(以干燥混凝土路面或沥青路面为例)

mu0 class-attribute instance-attribute

mu0: float = 0.8

滑转率为1时的摩擦系数

K class-attribute instance-attribute

K: float = 20

依赖于土壤黏聚系数和摩擦特性的常数

track_cmd_rate_limit class-attribute instance-attribute

track_cmd_rate_limit: float = 100

履带主动轮指令变化率限制 (rad/s^2)

min_throttle_manual class-attribute instance-attribute

min_throttle_manual: float = 0.08

MPC_MANTHR_MIN: Minimum manual thrust (0.0 - 1.0). Default: 0.08

min_throttle class-attribute instance-attribute

min_throttle: float = 0.12

MPC_THR_MIN: Minimum thrust in auto thrust control (0.05 - 1.0). Default: 0.12

max_throttle class-attribute instance-attribute

max_throttle: float = 1.0

MPC_THR_MAX: Maximum thrust in auto thrust control (0.0 - 1.0). Default: 1.0

hover_throttle class-attribute instance-attribute

hover_throttle: float = -1

MPC_THR_HOVER: Hover throttle manual (0.1 - 0.8). Default: 0.5 in PX4, calculated hover throttle in LAV2

__post_init__

__post_init__()

Calculate derived parameters.

RotorDynamics

RotorDynamics(params: VehicleParams = VehicleParams(), num_envs: int = 1, device: str | device = 'cpu')

Bases: DynamicsBase

Batched rotor dynamics model.

Initialize batched rotor dynamics on the specified device.

Methods:

Name	Description
randomize	Randomize rotor runtime parameters from nominal values.
update	Calculate the output thrust and torque for each motor based on the motor commands.
reset	Resets the rotor model to initial conditions.

Attributes:

Name	Type	Description
params	VehicleParams	Vehicle parameters for the dynamics model.

params instance-attribute

params: VehicleParams = params

Vehicle parameters for the dynamics model.

randomize

randomize(env_ids: slice | Sequence[int] | Tensor | None = None, randomization: dict[str, tuple[float, float]] | None = None) -> None

Randomize rotor runtime parameters from nominal values.

update

update(commands: Tensor) -> torch.Tensor

Calculate the output thrust and torque for each motor based on the motor commands.

Parameters:

Name	Type	Description	Default
commands	Tensor	Command tensor for each motor. Shape: (num_envs, 4)	required

Returns:

Type	Description
Tensor	torch.Tensor: Thrust and torque tensor for each motor. Shape: (num_envs, 8)

reset

reset(env_ids: slice | Sequence[int] | Tensor) -> None

Resets the rotor model to initial conditions.

Parameters:

Name	Type	Description	Default
env_ids	slice | Sequence[int] | Tensor	The environment ids to reset.	required

TrackDynamics

TrackDynamics(params: VehicleParams = VehicleParams(), num_envs: int = 1, device: str | device = 'cpu')

Bases: DynamicsBase

Batched track dynamics model.

Initialize batched track dynamics on the specified device.

Methods:

Name	Description
update	Compute track forces for batched environments.
reset	Resets the track model to initial conditions.

Attributes:

Name	Type	Description
params	VehicleParams	Vehicle parameters for the dynamics model.

params instance-attribute

params: VehicleParams = params

Vehicle parameters for the dynamics model.

update

update(commands: Tensor, state: Tensor) -> torch.Tensor

Compute track forces for batched environments.

Parameters:

Name	Type	Description	Default
commands	Tensor	Shape (num_envs, 2) track commands [left, right].	required
state	Tensor	Shape (num_envs, 3) body-frame velocities [u, v, r].	required

Returns:

Type	Description
Tensor	torch.Tensor: Tensor of shape (num_envs, 2, n_w, 2) [side, wheel, (Fx, Fy)].

reset

reset(env_ids: slice | Sequence[int] | Tensor) -> None

Resets the track model to initial conditions.

Parameters:

Name	Type	Description	Default
env_ids	slice | Sequence[int] | Tensor	The environment ids to reset.	required

ControlAction

ControlAction(cfg: ControlActionCfg, env: ManagerBasedRLEnv)

Bases: ActionTerm

Body torque control action term.

This action term applies a wrench to the drone body frame based on action commands

Attributes:

Name	Type	Description
cfg	ControlActionCfg	The configuration of the action term.

cfg instance-attribute

cfg: ControlActionCfg = cfg

The configuration of the action term.

ControlActionCfg

Bases: ActionTermCfg

See :class:ControlAction for more details.

Attributes:

Name	Type	Description
class_type	type[ActionTerm]	Class of the action term.
asset_name	str	Name of the asset in the environment for which the commands are generated.
params	VehicleParams	Vehicle parameters.
gains	dict | None	Optional gains configuration for the flight controller. See FlightController for details.
limits	dict | None	Optional command-limit configuration passed directly to the flight controller.
control_mode	str	Control mode, options: cmd_motor_thrusts, cmd_motor_speeds, cmd_ctbm, cmd_ctbr, cmd_ctatt, cmd_vel, cmd_pos.

class_type class-attribute instance-attribute

class_type: type[ActionTerm] = ControlAction

Class of the action term.

asset_name class-attribute instance-attribute

asset_name: str = 'robot'

Name of the asset in the environment for which the commands are generated.

params class-attribute instance-attribute

params: VehicleParams = VehicleParams()

Vehicle parameters.

gains class-attribute instance-attribute

gains: dict | None = None

Optional gains configuration for the flight controller. See FlightController for details.

limits class-attribute instance-attribute

limits: dict | None = None

Optional command-limit configuration passed directly to the flight controller.

control_mode class-attribute instance-attribute

control_mode: str = 'cmd_motor_speeds'

Control mode, options: cmd_motor_thrusts, cmd_motor_speeds, cmd_ctbm, cmd_ctbr, cmd_ctatt, cmd_vel, cmd_pos.

rotorpy ref: - https://github.com/spencerfolk/rotorpy/blob/main/rotorpy/learning/quadrotor_environments.py#L319C1-L350C12 - https://github.com/spencerfolk/rotorpy/blob/main/rotorpy/vehicles/multirotor.py#L997

TODO: also align control modes with px4, see: https://docs.px4.io/v1.14/en/flight_modes/offboard (version is unclear ATM)

TrackControlAction

TrackControlAction(cfg: TrackControlActionCfg, env: ManagerBasedRLEnv)

Bases: ActionTerm

Track force control action term.

This action term converts normalized track commands into ground reaction forces applied on each track body.

Attributes:

Name	Type	Description
cfg	TrackControlActionCfg	The configuration of the action term.

cfg instance-attribute

cfg: TrackControlActionCfg = cfg

The configuration of the action term.

TrackControlActionCfg

Bases: ActionTermCfg

See :class:ControlAction for more details.

Attributes:

Name	Type	Description
class_type	type[ActionTerm]	Class of the action term.
asset_name	str	Name of the asset in the environment for which the commands are generated.
params	VehicleParams	Vehicle parameters.

class_type class-attribute instance-attribute

class_type: type[ActionTerm] = TrackControlAction

Class of the action term.

asset_name class-attribute instance-attribute

asset_name: str = 'robot'

Name of the asset in the environment for which the commands are generated.

params class-attribute instance-attribute

params: VehicleParams = VehicleParams()

Vehicle parameters.

UniformPoseCommandGlobal

UniformPoseCommandGlobal(cfg: UniformPoseCommandGlobalCfg, env: ManagerBasedEnv)

Bases: CommandTerm

Command generator that generates pose commands containing a 3-D position and heading.

The command generator samples uniform 3D positions around the environment origin. The heading command is either set to point towards the target or is sampled uniformly. This can be configured through the :attr:UniformPoseCommandGlobalCfg.simple_heading parameter in the configuration.

Initialize the command generator class.

Parameters:

Name	Type	Description	Default
cfg	UniformPoseCommandGlobalCfg	The configuration parameters for the command generator.	required
env	ManagerBasedEnv	The environment object.	required

Attributes:

Name	Type	Description
cfg	UniformPoseCommandGlobalCfg	Configuration for the command generator.
command	Tensor	The desired 3D-pose in base frame. Shape is (num_envs, 4).

cfg instance-attribute

cfg: UniformPoseCommandGlobalCfg

Configuration for the command generator.

command property

command: Tensor

The desired 3D-pose in base frame. Shape is (num_envs, 4).

UniformPoseCommandGlobalCfg

Bases: CommandTermCfg

Configuration for the uniform 3D-pose command generator.

Classes:

Name	Description
Ranges	Uniform distribution ranges for the position commands.

Attributes:

Name	Type	Description
asset_name	str	Name of the asset in the environment for which the commands are generated.
simple_heading	bool	Whether to use simple heading or not.
ranges	Ranges	Distribution ranges for the position commands.
goal_pose_visualizer_cfg	VisualizationMarkersCfg	The configuration for the goal pose visualization marker. Defaults to GREEN_ARROW_X_MARKER_CFG.

asset_name class-attribute instance-attribute

asset_name: str = MISSING

Name of the asset in the environment for which the commands are generated.

simple_heading class-attribute instance-attribute

simple_heading: bool = MISSING

Whether to use simple heading or not.

If True, the heading is in the direction of the target position.

ranges class-attribute instance-attribute

ranges: Ranges = MISSING

Distribution ranges for the position commands.

goal_pose_visualizer_cfg class-attribute instance-attribute

goal_pose_visualizer_cfg: VisualizationMarkersCfg = replace(prim_path='/Visuals/Command/pose_goal')

The configuration for the goal pose visualization marker. Defaults to GREEN_ARROW_X_MARKER_CFG.

Ranges

Uniform distribution ranges for the position commands.

Attributes:

Name	Type	Description
pos_x	tuple[float, float]	Range for the x position (in m).
pos_y	tuple[float, float]	Range for the y position (in m).
pos_z	tuple[float, float]	Range for the z position (in m).
heading	tuple[float, float]	Heading range for the position commands (in rad).

pos_x class-attribute instance-attribute

pos_x: tuple[float, float] = MISSING

Range for the x position (in m).

pos_y class-attribute instance-attribute

pos_y: tuple[float, float] = MISSING

Range for the y position (in m).

pos_z class-attribute instance-attribute

pos_z: tuple[float, float] = MISSING

Range for the z position (in m).

heading class-attribute instance-attribute

heading: tuple[float, float] = MISSING

Heading range for the position commands (in rad).

Used only if :attr:simple_heading is False.

UniformVelocityGlobalCommand

UniformVelocityGlobalCommand(cfg: UniformVelocityGlobalCommandCfg, env: ManagerBasedEnv)

Bases: CommandTerm

Command generator that generates a velocity command with uniform distribution.

The command comprises linear velocity in x, y, z directions and an angular velocity around the z-axis. It is given in the robot's base frame.

If the :attr:cfg.heading_command flag is set to True, the angular velocity is computed from the heading error similar to doing a proportional control on the heading error. The target heading is sampled uniformly from the provided range. Otherwise, the angular velocity is sampled uniformly from the provided range.

Mathematically, the angular velocity is computed as follows from the heading command:

.. math::

\omega_z = \frac{1}{2} \text{wrap_to_pi}(\theta_{\text{target}} - \theta_{\text{current}})

Initialize the command generator.

Parameters:

Name	Type	Description	Default
cfg	UniformVelocityGlobalCommandCfg	The configuration of the command generator.	required
env	ManagerBasedEnv	The environment.	required

Raises:

Type	Description
ValueError	If the heading command is active but the heading range is not provided.

Methods:

Name	Description
__str__	Return a string representation of the command generator.

Attributes:

Name	Type	Description
cfg	UniformVelocityGlobalCommandCfg	The configuration of the command generator.
command	Tensor	The desired base velocity command in the base frame. Shape is (num_envs, 4).

cfg instance-attribute

cfg: UniformVelocityGlobalCommandCfg

The configuration of the command generator.

command property

command: Tensor

The desired base velocity command in the base frame. Shape is (num_envs, 4).

__str__

__str__() -> str

Return a string representation of the command generator.

UniformVelocityGlobalCommandCfg

Bases: CommandTermCfg

Configuration for the uniform velocity command generator.

Classes:

Name	Description
Ranges	Uniform distribution ranges for the velocity commands.

Attributes:

Name	Type	Description
asset_name	str	Name of the asset in the environment for which the commands are generated.
heading_command	bool	Whether to use heading command or angular velocity command. Defaults to False.
heading_control_stiffness	float	Scale factor to convert the heading error to angular velocity command. Defaults to 1.0.
rel_standing_envs	float	The sampled probability of environments that should be standing still. Defaults to 0.0.
rel_heading_envs	float	The sampled probability of environments where the robots follow the heading-based angular velocity command
ranges	Ranges	Distribution ranges for the velocity commands.
goal_vel_visualizer_cfg	VisualizationMarkersCfg	The configuration for the goal velocity visualization marker. Defaults to GREEN_ARROW_X_MARKER_CFG.
current_vel_visualizer_cfg	VisualizationMarkersCfg	The configuration for the current velocity visualization marker. Defaults to BLUE_ARROW_X_MARKER_CFG.

asset_name class-attribute instance-attribute

asset_name: str = MISSING

Name of the asset in the environment for which the commands are generated.

heading_command class-attribute instance-attribute

heading_command: bool = False

Whether to use heading command or angular velocity command. Defaults to False.

If True, the angular velocity command is computed from the heading error, where the target heading is sampled uniformly from provided range. Otherwise, the angular velocity command is sampled uniformly from provided range.

heading_control_stiffness class-attribute instance-attribute

heading_control_stiffness: float = 1.0

Scale factor to convert the heading error to angular velocity command. Defaults to 1.0.

rel_standing_envs class-attribute instance-attribute

rel_standing_envs: float = 0.0

The sampled probability of environments that should be standing still. Defaults to 0.0.

rel_heading_envs class-attribute instance-attribute

rel_heading_envs: float = 1.0

The sampled probability of environments where the robots follow the heading-based angular velocity command (the others follow the sampled angular velocity command). Defaults to 1.0.

This parameter is only used if :attr:heading_command is True.

ranges class-attribute instance-attribute

ranges: Ranges = MISSING

Distribution ranges for the velocity commands.

goal_vel_visualizer_cfg class-attribute instance-attribute

goal_vel_visualizer_cfg: VisualizationMarkersCfg = replace(prim_path='/Visuals/Command/velocity_goal')

The configuration for the goal velocity visualization marker. Defaults to GREEN_ARROW_X_MARKER_CFG.

current_vel_visualizer_cfg class-attribute instance-attribute

current_vel_visualizer_cfg: VisualizationMarkersCfg = replace(prim_path='/Visuals/Command/velocity_current')

The configuration for the current velocity visualization marker. Defaults to BLUE_ARROW_X_MARKER_CFG.

Ranges

Uniform distribution ranges for the velocity commands.

Attributes:

Name	Type	Description
lin_vel_x	tuple[float, float]	Range for the linear-x velocity command (in m/s).
lin_vel_y	tuple[float, float]	Range for the linear-y velocity command (in m/s).
lin_vel_z	tuple[float, float]	Range for the linear-z velocity command (in m/s).
ang_vel_z	tuple[float, float]	Range for the angular-z velocity command (in rad/s).
heading	tuple[float, float] | None	Range for the heading command (in rad). Defaults to None.

lin_vel_x class-attribute instance-attribute

lin_vel_x: tuple[float, float] = MISSING

Range for the linear-x velocity command (in m/s).

lin_vel_y class-attribute instance-attribute

lin_vel_y: tuple[float, float] = MISSING

Range for the linear-y velocity command (in m/s).

lin_vel_z class-attribute instance-attribute

lin_vel_z: tuple[float, float] = MISSING

Range for the linear-z velocity command (in m/s).

ang_vel_z class-attribute instance-attribute

ang_vel_z: tuple[float, float] = MISSING

Range for the angular-z velocity command (in rad/s).

heading class-attribute instance-attribute

heading: tuple[float, float] | None = None

Range for the heading command (in rad). Defaults to None.

This parameter is only used if :attr:~UniformVelocityGlobalCommandCfg.heading_command is True.

UniformBimodalVelocityCommand

UniformBimodalVelocityCommand(cfg: UniformBimodalVelocityCommandCfg, env: ManagerBasedEnv)

Bases: CommandTerm

Bimodal velocity command generator with a discrete mode indicator (4+1).

This command samples base-frame velocities with two modes: - Flight mode: uses ranges_flight - Ground mode: uses ranges_ground

The output command augments the 4-D velocity with a 1-D discrete indicator: - 0.0 for flight mode - 1.0 for ground/terrestrial mode

Heading behavior matches UniformVelocityGlobalCommand with optional heading-target proportional control. Ranges are applied per selected mode.

Attributes:

Name	Type	Description
command	Tensor	Return 4-D velocity plus 1-D mode indicator. Shape: (num_envs, 5).

command property

command: Tensor

Return 4-D velocity plus 1-D mode indicator. Shape: (num_envs, 5).

UniformBimodalVelocityCommandCfg

Bases: CommandTermCfg

Configuration for the bimodal velocity command generator (flight vs ground).

Classes:

Name	Description
Ranges	Uniform distribution ranges for velocity commands per mode.

Attributes:

Name	Type	Description
asset_name	str	Name of the asset receiving commands.
heading_command	bool	Whether to convert heading error to angular velocity (per-mode ranges apply).
heading_control_stiffness	float	Scale factor for heading error to yaw rate conversion.
rel_standing_envs	float	Probability that an environment should be standing still (zero command).
rel_heading_envs	float	Probability that an environment uses heading-based angular velocity when enabled.
rel_flight_envs	float	Probability of sampling flight mode (vs ground).
ranges_flight	Ranges	Ranges for flight mode sampling.
ranges_ground	Ranges	Ranges for ground/terrestrial mode sampling.
goal_vel_visualizer_cfg	VisualizationMarkersCfg	The configuration for the goal velocity visualization marker. Defaults to GREEN_ARROW_X_MARKER_CFG.
current_vel_visualizer_cfg	VisualizationMarkersCfg	The configuration for the current velocity visualization marker. Defaults to BLUE_ARROW_X_MARKER_CFG.

asset_name class-attribute instance-attribute

asset_name: str = MISSING

Name of the asset receiving commands.

heading_command class-attribute instance-attribute

heading_command: bool = False

Whether to convert heading error to angular velocity (per-mode ranges apply).

heading_control_stiffness class-attribute instance-attribute

heading_control_stiffness: float = 1.0

Scale factor for heading error to yaw rate conversion.

rel_standing_envs class-attribute instance-attribute

rel_standing_envs: float = 0.0

Probability that an environment should be standing still (zero command).

rel_heading_envs class-attribute instance-attribute

rel_heading_envs: float = 1.0

Probability that an environment uses heading-based angular velocity when enabled.

rel_flight_envs class-attribute instance-attribute

rel_flight_envs: float = 0.5

Probability of sampling flight mode (vs ground).

ranges_flight class-attribute instance-attribute

ranges_flight: Ranges = MISSING

Ranges for flight mode sampling.

ranges_ground class-attribute instance-attribute

ranges_ground: Ranges = MISSING

Ranges for ground/terrestrial mode sampling.

goal_vel_visualizer_cfg class-attribute instance-attribute

goal_vel_visualizer_cfg: VisualizationMarkersCfg = VisualizationMarkersCfg(prim_path='/Visuals/Command/velocity_goal', markers={'flight': UsdFileCfg(usd_path=f'{ISAAC_NUCLEUS_DIR}/Props/UIElements/arrow_x.usd', scale=(1.0, 0.1, 0.1), visual_material=PreviewSurfaceCfg(diffuse_color=(0.0, 1.0, 0.0))), 'ground': UsdFileCfg(usd_path=f'{ISAAC_NUCLEUS_DIR}/Props/UIElements/arrow_x.usd', scale=(1.0, 0.1, 0.1), visual_material=PreviewSurfaceCfg(diffuse_color=(1.0, 0.0, 0.0)))})

The configuration for the goal velocity visualization marker. Defaults to GREEN_ARROW_X_MARKER_CFG.

current_vel_visualizer_cfg class-attribute instance-attribute

current_vel_visualizer_cfg: VisualizationMarkersCfg = replace(prim_path='/Visuals/Command/velocity_current')

The configuration for the current velocity visualization marker. Defaults to BLUE_ARROW_X_MARKER_CFG.

Ranges

Uniform distribution ranges for velocity commands per mode.

Attributes:

Name	Type	Description
lin_vel_x	tuple[float, float]	Range for the linear-x velocity command (in m/s).
lin_vel_y	tuple[float, float]	Range for the linear-y velocity command (in m/s).
lin_vel_z	tuple[float, float]	Range for the linear-z velocity command (in m/s).
ang_vel_z	tuple[float, float]	Range for the angular-z velocity command (in rad/s).
heading	tuple[float, float] | None	Range for the heading command (in rad). Defaults to None.

lin_vel_x class-attribute instance-attribute

lin_vel_x: tuple[float, float] = MISSING

Range for the linear-x velocity command (in m/s).

lin_vel_y class-attribute instance-attribute

lin_vel_y: tuple[float, float] = MISSING

Range for the linear-y velocity command (in m/s).

lin_vel_z class-attribute instance-attribute

lin_vel_z: tuple[float, float] = MISSING

Range for the linear-z velocity command (in m/s).

ang_vel_z class-attribute instance-attribute

ang_vel_z: tuple[float, float] = MISSING

Range for the angular-z velocity command (in rad/s).

heading class-attribute instance-attribute

heading: tuple[float, float] | None = None

Range for the heading command (in rad). Defaults to None.

This parameter is only used if :attr:~UniformBimodalVelocityCommandCfg.heading_command is True.

HelixTrajectory dataclass

HelixTrajectory(*, yaw_from_velocity: bool = True, fixed_yaw: Tensor | None = None, attitude: Tensor | None = None, body_rate: Tensor | None = None, center: Tensor, radius: Tensor, omega: Tensor, z_amplitude: Tensor, z_frequency: Tensor, z_phase: Tensor)

Bases: SpatialTrajectory

Batched helix trajectory with optional vertical oscillation.

Methods:

Name	Description
sample_heading	Sample batched yaw and yaw-rate references for time values t.
build_target	Assemble a batched 12D target tensor from motion primitives.
__post_init__	Initialize shared tensors from the center batch shape when available.
sample	Sample the batched helix trajectory at time values t.

sample_heading

sample_heading(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> tuple[torch.Tensor, torch.Tensor]

Sample batched yaw and yaw-rate references for time values t.

build_target

build_target(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> torch.Tensor

Assemble a batched 12D target tensor from motion primitives.

__post_init__

__post_init__()

Initialize shared tensors from the center batch shape when available.

sample

sample(t: Tensor) -> torch.Tensor

Sample the batched helix trajectory at time values t.

LemniscateTrajectory dataclass

LemniscateTrajectory(*, yaw_from_velocity: bool = True, fixed_yaw: Tensor | None = None, attitude: Tensor | None = None, body_rate: Tensor | None = None, center: Tensor, scale: Tensor, omega: Tensor, z_amplitude: Tensor, z_frequency: Tensor, z_phase: Tensor)

Bases: SpatialTrajectory

Batched Bernoulli lemniscate trajectory with vertical oscillation.

Methods:

Name	Description
sample_heading	Sample batched yaw and yaw-rate references for time values t.
build_target	Assemble a batched 12D target tensor from motion primitives.
__post_init__	Initialize shared tensors from the center batch shape when available.
sample	Sample the batched lemniscate trajectory at time values t.

sample_heading

sample_heading(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> tuple[torch.Tensor, torch.Tensor]

Sample batched yaw and yaw-rate references for time values t.

build_target

build_target(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> torch.Tensor

Assemble a batched 12D target tensor from motion primitives.

__post_init__

__post_init__()

Initialize shared tensors from the center batch shape when available.

sample

sample(t: Tensor) -> torch.Tensor

Sample the batched lemniscate trajectory at time values t.

LissajousTrajectory dataclass

LissajousTrajectory(*, yaw_from_velocity: bool = True, fixed_yaw: Tensor | None = None, attitude: Tensor | None = None, body_rate: Tensor | None = None, center: Tensor, amplitude: Tensor, frequency: Tensor, phase: Tensor, omega: Tensor)

Bases: SpatialTrajectory

Batched three-dimensional Lissajous trajectory.

Methods:

Name	Description
sample_heading	Sample batched yaw and yaw-rate references for time values t.
build_target	Assemble a batched 12D target tensor from motion primitives.
__post_init__	Initialize shared tensors from the center batch shape when available.
sample	Sample the batched Lissajous trajectory at time values t.

sample_heading

sample_heading(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> tuple[torch.Tensor, torch.Tensor]

Sample batched yaw and yaw-rate references for time values t.

build_target

build_target(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> torch.Tensor

Assemble a batched 12D target tensor from motion primitives.

__post_init__

__post_init__()

Initialize shared tensors from the center batch shape when available.

sample

sample(t: Tensor) -> torch.Tensor

Sample the batched Lissajous trajectory at time values t.

RectangleTrajectory dataclass

RectangleTrajectory(*, yaw_from_velocity: bool = True, fixed_yaw: Tensor | None = None, attitude: Tensor | None = None, body_rate: Tensor | None = None, center: Tensor, size: Tensor, speed: Tensor)

Bases: SpatialTrajectory

Batched constant-speed rectangle trajectory in the horizontal plane.

Methods:

Name	Description
sample_heading	Sample batched yaw and yaw-rate references for time values t.
build_target	Assemble a batched 12D target tensor from motion primitives.
__post_init__	Initialize shared tensors from the center batch shape when available.
sample	Sample the batched rectangle trajectory at time values t.

sample_heading

sample_heading(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> tuple[torch.Tensor, torch.Tensor]

Sample batched yaw and yaw-rate references for time values t.

build_target

build_target(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> torch.Tensor

Assemble a batched 12D target tensor from motion primitives.

__post_init__

__post_init__()

Initialize shared tensors from the center batch shape when available.

sample

sample(t: Tensor) -> torch.Tensor

Sample the batched rectangle trajectory at time values t.

RandomTrajectoryCommand

RandomTrajectoryCommand(cfg: RandomTrajectoryCommandCfg, env: ManagerBasedEnv)

Bases: CommandTerm

Trajectory command generator with randomized type and parameters per environment.

RandomTrajectoryCommandCfg

Bases: CommandTermCfg

Configuration for randomized trajectory commands.

randomize_action_term_attr

randomize_action_term_attr(env, env_ids: Tensor | None, action_term_name: str = 'control_action', attr_name: str = '_flight_controller', **kwargs)

Randomize a selected action-term attribute.

randomize_controller_params

randomize_controller_params(env, env_ids: Tensor | None, action_term_name: str = 'control_action', controller_attr: str = '_flight_controller', gains: dict | None = None, limits: dict | None = None)

Randomize controller gains and command limits attached to an action term.

randomize_mixer_params

randomize_mixer_params(env, env_ids: Tensor | None, action_term_name: str = 'control_action', mixer_attr: str = '_mixer', param_range: dict[str, tuple[float, float]] | None = None)

Randomize mixer parameters attached to an action term.

randomize_rotor_params

randomize_rotor_params(env, env_ids: Tensor | None, action_term_name: str = 'control_action', rotor_attr: str = '_rotor_dynamics', param_range: dict[str, tuple[float, float]] | None = None)

Randomize rotor parameters attached to an action term.

target_pos_b

target_pos_b(env: ManagerBasedRLEnv, command_name: str | None = None, asset_cfg: SceneEntityCfg = SceneEntityCfg('robot')) -> torch.Tensor

Position of target in body frame.

root_euler_w

root_euler_w(env: ManagerBasedRLEnv, asset_cfg: SceneEntityCfg = SceneEntityCfg('robot')) -> torch.Tensor

Euler angles of the root in world frame.

lin_vel_l2

lin_vel_l2(env: ManagerBasedRLEnv, asset_cfg: SceneEntityCfg = SceneEntityCfg('robot')) -> torch.Tensor

Penalize base linear velocity using L2 squared kernel.

ang_vel_l2

ang_vel_l2(env: ManagerBasedRLEnv, asset_cfg: SceneEntityCfg = SceneEntityCfg('robot')) -> torch.Tensor

Penalize base angular velocity using L2 squared kernel.

pos_error_l2

pos_error_l2(env: ManagerBasedRLEnv, command_name: str) -> torch.Tensor

Penalize asset pos from its target pos using L2 squared kernel.

pos_error_tanh

pos_error_tanh(env: ManagerBasedRLEnv, std: float, command_name: str) -> torch.Tensor

Penalize asset pos from its target pos using tanh kernel.

yaw_error_l2

yaw_error_l2(env: ManagerBasedRLEnv, command_name: str) -> torch.Tensor

Penalize heading error from target heading using L2 squared kernel.

yaw_error_tanh

yaw_error_tanh(env: ManagerBasedRLEnv, std: float, command_name: str) -> torch.Tensor

Penalize heading error from target heading using tanh kernel.

track_lin_vel_z_exp

track_lin_vel_z_exp(env: ManagerBasedRLEnv, std: float, command_name: str, is_bimodal: bool = False, asset_cfg: SceneEntityCfg = SceneEntityCfg('robot')) -> torch.Tensor

Reward tracking of linear velocity commands (z axis) using exponential kernel.

track_lin_vel_exp

track_lin_vel_exp(env: ManagerBasedRLEnv, std: float, command_name: str, asset_cfg: SceneEntityCfg = SceneEntityCfg('robot')) -> torch.Tensor

Reward tracking of linear velocity commands using exponential kernel.

track_yaw_vel_exp

track_yaw_vel_exp(env: ManagerBasedRLEnv, std: float, command_name: str, asset_cfg: SceneEntityCfg = SceneEntityCfg('robot')) -> torch.Tensor

Reward tracking of angular velocity commands (yaw) using exponential kernel.

bimodal_action_tanh

bimodal_action_tanh(env: ManagerBasedRLEnv, std: float, command_name: str, flight_action_name: str = 'control_action', ground_action_name: str = 'track_control_action', ground_weight: float = 1.0, flight_weight: float = 1.0) -> torch.Tensor

Penalize bimodal actions using tanh kernel.

bimodal_height_tanh

bimodal_height_tanh(env: ManagerBasedRLEnv, std: float, command_name: str, asset_cfg: SceneEntityCfg = SceneEntityCfg('robot'), ground_weight: float = 1.0, flight_weight: float = 1.0) -> torch.Tensor

Penalize bimodal height using tanh kernel.

contact_impulse

contact_impulse(env: ManagerBasedRLEnv, threshold: float, sensor_cfg: SceneEntityCfg, mode: str = 'threshold') -> torch.Tensor

Penalize excessive contact impulse (rate of change of contact forces).

This function calculates the impulse as the change in contact forces between consecutive time steps and penalizes values that exceed a threshold.

Parameters:

Name	Type	Description	Default
env	ManagerBasedRLEnv	The learning environment.	required
threshold	float	Maximum acceptable impulse magnitude. Forces below this are not penalized.	required
sensor_cfg	SceneEntityCfg	Configuration for the contact sensor, specifying which bodies to monitor.	required
mode	str	Penalty calculation mode: - "threshold": Penalize only the amount exceeding threshold (continuous) - "binary": Return 1.0 if any impulse exceeds threshold, else 0.0 (discrete) - "total": Return total impulse magnitude regardless of threshold (for monitoring)	'threshold'

Returns:

Type	Description
Tensor	torch.Tensor: Penalty value for each environment: - "threshold" mode: Sum of (impulse - threshold) for all violations - "binary" mode: 1.0 if violation exists, 0.0 otherwise - "total" mode: Total impulse magnitude

Raises:

Type	Description
ValueError	If the sensor history length is less than 2 or the mode is invalid.

Examples:

>>> # Penalize hard landings (impulse > 50 N·s)
>>> impulse_penalty = contact_impulse(
...     env, threshold=50.0,
...     sensor_cfg=SceneEntityCfg("contact_sensor", body_ids=[arml_id, armr_id]),
...     mode="threshold"
... )
>>> rewards["impulse_penalty"] = impulse_penalty * -1.0

bimodal_contacts

bimodal_contacts(env: ManagerBasedRLEnv, command_name: str, threshold: float, sensor_cfg: SceneEntityCfg, mode: str = 'threshold', ground_weight: float = 1.0, flight_weight: float = 1.0) -> torch.Tensor

Penalize contacts when switching from flight to ground mode.

died

died(env: ManagerBasedRLEnv, asset_cfg: SceneEntityCfg = SceneEntityCfg('robot'), min: float = 0.2, max: float = 2.0) -> torch.Tensor

Termination condition based on the robot's height.

commands

Classes:

Name	Description
RandomTrajectoryCommand	Trajectory command generator with randomized type and parameters per environment.
RandomTrajectoryCommandCfg	Configuration for randomized trajectory commands.

RandomTrajectoryCommand

RandomTrajectoryCommand(cfg: RandomTrajectoryCommandCfg, env: ManagerBasedEnv)

Bases: CommandTerm

Trajectory command generator with randomized type and parameters per environment.

RandomTrajectoryCommandCfg

Bases: CommandTermCfg

Configuration for randomized trajectory commands.

mjlab

Modules:

Name	Description
LAV2_base

LAV2_base

Modules:

Name	Description
LAV2_rl_cfg
mdp

LAV2_rl_cfg

Functions:

Name	Description
create_LAV2_ppo_runner_cfg	Create RL runner configuration for LAV2 base task.

create_LAV2_ppo_runner_cfg

create_LAV2_ppo_runner_cfg() -> RslRlOnPolicyRunnerCfg

Create RL runner configuration for LAV2 base task.

mdp

Modules:

Name	Description
actions
commands
rewards

Classes:

Name	Description
FlightActionMapper	Map batched normalized actions into controller targets, wrench, or RPM.
Mixer	Map thrust and moment demands to rotor commands.
VehicleParams	LAV2 vehicle parameters.
RotorDynamics	Batched rotor dynamics model.
ControlAction	Body torque control action term.
ControlActionCfg	See :class:ControlAction for more details.
UniformPoseCommandGlobal	Command generator that generates pose commands containing a 3-D position and heading.
UniformPoseCommandGlobalCfg	Configuration for the uniform 3D-pose command generator.

Functions:

Name	Description
lin_vel_l2	Penalize base linear velocity using L2 squared kernel.
ang_vel_l2	Penalize base angular velocity using L2 squared kernel.
pos_error_l2	Penalize asset pos from its target pos using L2 squared kernel.
pos_error_tanh	Penalize asset pos from its target pos using tanh kernel.
yaw_error_l2	Penalize heading error from target heading using L2 squared kernel.
yaw_error_tanh	Penalize heading error from target heading using tanh kernel.

FlightActionMapper

FlightActionMapper(control_mode: str, params: VehicleParams = VehicleParams(), limits: FlightMappingLimits = FlightMappingLimits(), *, num_envs: int = 1, device: str | device = 'cpu')

Map batched normalized actions into controller targets, wrench, or RPM.

Initialize a batched Torch flight action mapper for the selected mode.

Methods:

Name	Description
map_action	Map a normalized batched action into RPM commands and intermediates.

Attributes:

Name	Type	Description
requires_controller	bool	Whether this mode needs a controller target expansion step.

requires_controller property

requires_controller: bool

Whether this mode needs a controller target expansion step.

map_action

map_action(action: Tensor, mixer: Mixer, *, state: Tensor | None = None, flight_controller: ControllerBase | None = None) -> FlightMappingResult

Map a normalized batched action into RPM commands and intermediates.

Mixer

Mixer(params: VehicleParams = VehicleParams(), num_envs: int = 1, device: str | device = 'cpu')

Map thrust and moment demands to rotor commands.

Note

电机推力分配： 电机1 = 推力 - 横滚(roll) + 俯仰(pitch) + 偏航(yaw) 电机2 = 推力 + 横滚(roll) - 俯仰(pitch) + 偏航(yaw) 电机3 = 推力 + 横滚(roll) + 俯仰(pitch) - 偏航(yaw) 电机4 = 推力 - 横滚(roll) - 俯仰(pitch) - 偏航(yaw)

Initialize the torch mixer with vehicle parameters and device.

Methods:

Name	Description
randomize	Randomize mixer runtime parameters from nominal values.
apply_thrust_curve	Apply thrust curve to normalized thrust command.
calculate_rotor_commands	Allocate thrust and moments into rotor speeds.

randomize

randomize(env_ids=None, randomization=None)

Randomize mixer runtime parameters from nominal values.

apply_thrust_curve

apply_thrust_curve(normalized_thrust: Tensor, mode: int = 1) -> torch.Tensor

Apply thrust curve to normalized thrust command.

References

cmd_ctatt: https://github.com/PX4/PX4-Autopilot/blob/e5a483e4b71ab4b994cb0121fe28992d8ce0c8cd/src/modules/mc_att_control/mc_att_control_main.cpp#L111-L137 cmd_ctbr: https://github.com/PX4/PX4-Autopilot/blob/e5a483e4b71ab4b994cb0121fe28992d8ce0c8cd/src/modules/mc_rate_control/MulticopterRateControl.cpp#L156-L179

Parameters:

Name	Type	Description	Default
normalized_thrust	Tensor	Normalized thrust command [-1, 1]. Shape: arbitrary, e.g. (num_envs, n_motors). Applied elementwise.	required
mode	int	Thrust curve mode.	1

Returns:

Type	Description
Tensor	torch.Tensor: Thrust in Newtons with the same shape as normalized_thrust.

Raises:

Type	Description
NotImplementedError	If the specified mode is not implemented.

calculate_rotor_commands

calculate_rotor_commands(control: Tensor) -> torch.Tensor

Allocate thrust and moments into rotor speeds.

Parameters:

Name	Type	Description	Default
control	Tensor	Desired total thrust and roll/pitch/yaw moments. Shape: (num_envs, 4)	required

Returns:

Type	Description
Tensor	torch.Tensor: The amplitude of each motor's rotational speed (calculated using sqrt(clip(rpm_sq))). Shape: (num_envs, 4)

VehicleParams dataclass

VehicleParams(rho: float = 1.184, g: float = 9.81, sim_dt: float = 0.01, step_dt: float = 0.01, randomization: dict[str, dict[str, tuple[float, float]]] = dict(), mass: float = 2.1, inertia: list = (lambda: np.diag([0.11, 0.108, 0.215]).tolist())(), diameter: float = 0.476, Cdx: float = 0.5, x: float = 0.0952, y: float = 0.114423, h: float = -0.0125, alpha: float = 8 * (np.pi / 180), r_p: float = 0.0775, J_m: float = 1e-05, Ct: float = 0.666, Cq: float = 0.0716, theta0: float = 14.6 * (np.pi / 180), thetaTip: float = 6.8 * (np.pi / 180), lock: float = 0.6051, max_rpm: float = 10000.0, init_rpm: float = -1.0, rotor_rpm_rate_limit: float = 50000.0, tau_f: float = 0.01, tau_m: float = 0.05, tau_up: float = -1.0, tau_down: float = -1.0, r_w: float = 0.0285, n_w: int = 4, B: float = 0.136, L: float = 0.23, h_cg: float = 0.078, f: float = 0.01, mu0: float = 0.8, K: float = 20, track_cmd_rate_limit: float = 100, min_throttle_manual: float = 0.08, min_throttle: float = 0.12, max_throttle: float = 1.0, hover_throttle: float = -1)

LAV2 vehicle parameters.

Methods:

Name	Description
__post_init__	Calculate derived parameters.

Attributes:

Name	Type	Description
rho	float	空气密度 (kg/m^3)
g	float	重力加速度 (m/s^2)
sim_dt	float	仿真步长 (s)
step_dt	float	控制步长 (s)
randomization	dict[str, dict[str, tuple[float, float]]]	运行时参数随机化配置，按组件 (e.g. rotor/mixer) 划分为多个 scale 相对范围子表
mass	float	质量 (kg)
inertia	list	惯性矩阵 (kg*m^2)
diameter	float	直径 (m)
Cdx	float	空气阻力系数
x	float	x方向臂长 (m)
y	float	y方向臂长 (m)
h	float	重心高度 (m)
alpha	float	y方向涵道外倾角度 (rad)
r_p	float	螺旋桨半径 (m)
J_m	float	电机转动惯量 (kg*m^2)
Ct	float	推力系数
Cq	float	扭矩系数
theta0	float	桨根角 (rad)
thetaTip	float	桨尖角 (rad)
theta1	float	桨叶扭转角 (rad)
lock	float	Lock number
max_rpm	float	最大转速
init_rpm	float	初始转速
rotor_rpm_rate_limit	float	旋翼转速变化率限制 (rpm/s)
cT	float	转速平方到推力增益 (N/(rpm^2))
cM	float	转速平方到扭矩增益 (N*m/(rpm^2))
tau_f	float	电机滤波时间常数 (s)
tau_m	float	兼容保留的电机时间常数 (s)
tau_up	float	电机升速时间常数 (s)
tau_down	float	电机降速时间常数 (s)
r_w	float	主动轮半径 (m)
n_w	int	负重轮个数
B	float	两侧履带中心距 (m)
L	float	履带节点接地部分长度 (m)
h_cg	float	车体质心高度 (m)
f	float	滚动阻力系数(以干燥混凝土路面或沥青路面为例)
mu0	float	滑转率为1时的摩擦系数
K	float	依赖于土壤黏聚系数和摩擦特性的常数
track_cmd_rate_limit	float	履带主动轮指令变化率限制 (rad/s^2)
min_throttle_manual	float	MPC_MANTHR_MIN: Minimum manual thrust (0.0 - 1.0). Default: 0.08
min_throttle	float	MPC_THR_MIN: Minimum thrust in auto thrust control (0.05 - 1.0). Default: 0.12
max_throttle	float	MPC_THR_MAX: Maximum thrust in auto thrust control (0.0 - 1.0). Default: 1.0
hover_throttle	float	MPC_THR_HOVER: Hover throttle manual (0.1 - 0.8). Default: 0.5 in PX4, calculated hover throttle in LAV2

rho class-attribute instance-attribute

rho: float = 1.184

空气密度 (kg/m^3)

g class-attribute instance-attribute

g: float = 9.81

重力加速度 (m/s^2)

sim_dt class-attribute instance-attribute

sim_dt: float = 0.01

仿真步长 (s)

step_dt class-attribute instance-attribute

step_dt: float = 0.01

控制步长 (s)

randomization class-attribute instance-attribute

randomization: dict[str, dict[str, tuple[float, float]]] = field(default_factory=dict)

运行时参数随机化配置，按组件 (e.g. rotor/mixer) 划分为多个 scale 相对范围子表

mass class-attribute instance-attribute

mass: float = 2.1

质量 (kg)

inertia class-attribute instance-attribute

inertia: list = field(default_factory=lambda: tolist())

惯性矩阵 (kg*m^2)

diameter class-attribute instance-attribute

diameter: float = 0.476

直径 (m)

Cdx class-attribute instance-attribute

Cdx: float = 0.5

空气阻力系数

x class-attribute instance-attribute

x: float = 0.0952

x方向臂长 (m)

y class-attribute instance-attribute

y: float = 0.114423

y方向臂长 (m)

h class-attribute instance-attribute

h: float = -0.0125

重心高度 (m)

alpha class-attribute instance-attribute

alpha: float = 8 * (pi / 180)

y方向涵道外倾角度 (rad)

r_p class-attribute instance-attribute

r_p: float = 0.0775

螺旋桨半径 (m)

J_m class-attribute instance-attribute

J_m: float = 1e-05

电机转动惯量 (kg*m^2)

Ct class-attribute instance-attribute

Ct: float = 0.666

推力系数

Cq class-attribute instance-attribute

Cq: float = 0.0716

扭矩系数

theta0 class-attribute instance-attribute

theta0: float = 14.6 * (pi / 180)

桨根角 (rad)

thetaTip class-attribute instance-attribute

thetaTip: float = 6.8 * (pi / 180)

桨尖角 (rad)

theta1 class-attribute instance-attribute

theta1: float = field(init=False)

桨叶扭转角 (rad)

lock class-attribute instance-attribute

lock: float = 0.6051

Lock number

max_rpm class-attribute instance-attribute

max_rpm: float = 10000.0

最大转速

init_rpm class-attribute instance-attribute

init_rpm: float = -1.0

初始转速

rotor_rpm_rate_limit class-attribute instance-attribute

rotor_rpm_rate_limit: float = 50000.0

旋翼转速变化率限制 (rpm/s)

cT class-attribute instance-attribute

cT: float = field(init=False)

转速平方到推力增益 (N/(rpm^2))

cM class-attribute instance-attribute

cM: float = field(init=False)

转速平方到扭矩增益 (N*m/(rpm^2))

tau_f class-attribute instance-attribute

tau_f: float = 0.01

电机滤波时间常数 (s)

tau_m class-attribute instance-attribute

tau_m: float = 0.05

兼容保留的电机时间常数 (s)

tau_up class-attribute instance-attribute

tau_up: float = -1.0

电机升速时间常数 (s)

tau_down class-attribute instance-attribute

tau_down: float = -1.0

电机降速时间常数 (s)

r_w class-attribute instance-attribute

r_w: float = 0.0285

主动轮半径 (m)

n_w class-attribute instance-attribute

n_w: int = 4

负重轮个数

B class-attribute instance-attribute

B: float = 0.136

两侧履带中心距 (m)

L class-attribute instance-attribute

L: float = 0.23

履带节点接地部分长度 (m)

h_cg class-attribute instance-attribute

h_cg: float = 0.078

车体质心高度 (m)

f class-attribute instance-attribute

f: float = 0.01

滚动阻力系数(以干燥混凝土路面或沥青路面为例)

mu0 class-attribute instance-attribute

mu0: float = 0.8

滑转率为1时的摩擦系数

K class-attribute instance-attribute

K: float = 20

依赖于土壤黏聚系数和摩擦特性的常数

track_cmd_rate_limit class-attribute instance-attribute

track_cmd_rate_limit: float = 100

履带主动轮指令变化率限制 (rad/s^2)

min_throttle_manual class-attribute instance-attribute

min_throttle_manual: float = 0.08

MPC_MANTHR_MIN: Minimum manual thrust (0.0 - 1.0). Default: 0.08

min_throttle class-attribute instance-attribute

min_throttle: float = 0.12

MPC_THR_MIN: Minimum thrust in auto thrust control (0.05 - 1.0). Default: 0.12

max_throttle class-attribute instance-attribute

max_throttle: float = 1.0

MPC_THR_MAX: Maximum thrust in auto thrust control (0.0 - 1.0). Default: 1.0

hover_throttle class-attribute instance-attribute

hover_throttle: float = -1

MPC_THR_HOVER: Hover throttle manual (0.1 - 0.8). Default: 0.5 in PX4, calculated hover throttle in LAV2

__post_init__

__post_init__()

Calculate derived parameters.

RotorDynamics

RotorDynamics(params: VehicleParams = VehicleParams(), num_envs: int = 1, device: str | device = 'cpu')

Bases: DynamicsBase

Batched rotor dynamics model.

Initialize batched rotor dynamics on the specified device.

Methods:

Name	Description
randomize	Randomize rotor runtime parameters from nominal values.
update	Calculate the output thrust and torque for each motor based on the motor commands.
reset	Resets the rotor model to initial conditions.

Attributes:

Name	Type	Description
params	VehicleParams	Vehicle parameters for the dynamics model.

params instance-attribute

params: VehicleParams = params

Vehicle parameters for the dynamics model.

randomize

randomize(env_ids: slice | Sequence[int] | Tensor | None = None, randomization: dict[str, tuple[float, float]] | None = None) -> None

Randomize rotor runtime parameters from nominal values.

update

update(commands: Tensor) -> torch.Tensor

Calculate the output thrust and torque for each motor based on the motor commands.

Parameters:

Name	Type	Description	Default
commands	Tensor	Command tensor for each motor. Shape: (num_envs, 4)	required

Returns:

Type	Description
Tensor	torch.Tensor: Thrust and torque tensor for each motor. Shape: (num_envs, 8)

reset

reset(env_ids: slice | Sequence[int] | Tensor) -> None

Resets the rotor model to initial conditions.

Parameters:

Name	Type	Description	Default
env_ids	slice | Sequence[int] | Tensor	The environment ids to reset.	required

ControlAction

ControlAction(cfg: ControlActionCfg, env: ManagerBasedRlEnv)

Bases: ActionTerm

Body torque control action term.

This action term applies a wrench to the drone body frame based on action commands

ControlActionCfg dataclass

ControlActionCfg(*, class_type: type[ActionTerm] = ControlAction, entity_name: str = 'robot', params: VehicleParams = VehicleParams())

Bases: ActionTermCfg

See :class:ControlAction for more details.

Attributes:

Name	Type	Description
class_type	type[ActionTerm]	Class of the action term.
entity_name	str	Name of the asset in the environment for which the commands are generated.
params	VehicleParams	Vehicle parameters.

class_type class-attribute instance-attribute

class_type: type[ActionTerm] = ControlAction

Class of the action term.

entity_name class-attribute instance-attribute

entity_name: str = 'robot'

Name of the asset in the environment for which the commands are generated.

params class-attribute instance-attribute

params: VehicleParams = field(default_factory=VehicleParams)

Vehicle parameters.

UniformPoseCommandGlobal

UniformPoseCommandGlobal(cfg: UniformPoseCommandGlobalCfg, env: ManagerBasedRlEnv)

Bases: CommandTerm

Command generator that generates pose commands containing a 3-D position and heading. The command generator samples uniform 3D positions around the environment origin. The heading command is either set to point towards the target or is sampled uniformly. This can be configured through the :attr:UniformPoseCommandGlobalCfg.simple_heading parameter in the configuration.

Initialize the command generator class.

Parameters:

Name	Type	Description	Default
cfg	UniformPoseCommandGlobalCfg	The configuration parameters for the command generator.	required
env	ManagerBasedRlEnv	The environment object.	required

Attributes:

Name	Type	Description
cfg	UniformPoseCommandGlobalCfg	Configuration for the command generator.
command	Tensor	The desired 3D-pose in base frame. Shape is (num_envs, 4).

cfg instance-attribute

cfg: UniformPoseCommandGlobalCfg

Configuration for the command generator.

command property

command: Tensor

The desired 3D-pose in base frame. Shape is (num_envs, 4).

UniformPoseCommandGlobalCfg dataclass

UniformPoseCommandGlobalCfg(*, class_type: type = UniformPoseCommandGlobal, entity_name: str, simple_heading: bool, ranges: Ranges, viz: VizCfg = VizCfg())

Bases: CommandTermCfg

Configuration for the uniform 3D-pose command generator.

Classes:

Name	Description
Ranges	Uniform distribution ranges for the position commands.

Attributes:

Name	Type	Description
entity_name	str	Name of the asset in the environment for which the commands are generated.
simple_heading	bool	Whether to use simple heading or not.
ranges	Ranges	Distribution ranges for the position commands.

entity_name instance-attribute

entity_name: str

Name of the asset in the environment for which the commands are generated.

simple_heading instance-attribute

simple_heading: bool

Whether to use simple heading or not.

If True, the heading is in the direction of the target position.

ranges instance-attribute

ranges: Ranges

Distribution ranges for the position commands.

Ranges dataclass

Ranges(pos_x: tuple[float, float], pos_y: tuple[float, float], pos_z: tuple[float, float], heading: tuple[float, float])

Uniform distribution ranges for the position commands.

Attributes:

Name	Type	Description
pos_x	tuple[float, float]	Range for the x position (in m).
pos_y	tuple[float, float]	Range for the y position (in m).
pos_z	tuple[float, float]	Range for the z position (in m).
heading	tuple[float, float]	Heading range for the position commands (in rad).

pos_x instance-attribute

pos_x: tuple[float, float]

Range for the x position (in m).

pos_y instance-attribute

pos_y: tuple[float, float]

Range for the y position (in m).

pos_z instance-attribute

pos_z: tuple[float, float]

Range for the z position (in m).

heading instance-attribute

heading: tuple[float, float]

Heading range for the position commands (in rad).

Used only if :attr:simple_heading is False.

lin_vel_l2

lin_vel_l2(env: ManagerBasedRlEnv, asset_cfg: SceneEntityCfg = _DEFAULT_ASSET_CFG) -> torch.Tensor

Penalize base linear velocity using L2 squared kernel.

ang_vel_l2

ang_vel_l2(env: ManagerBasedRlEnv, asset_cfg: SceneEntityCfg = _DEFAULT_ASSET_CFG) -> torch.Tensor

Penalize base angular velocity using L2 squared kernel.

pos_error_l2

pos_error_l2(env: ManagerBasedRlEnv, command_name: str, asset_cfg: SceneEntityCfg = _DEFAULT_ASSET_CFG) -> torch.Tensor

Penalize asset pos from its target pos using L2 squared kernel.

pos_error_tanh

pos_error_tanh(env: ManagerBasedRlEnv, command_name: str, std: float = 0.8, asset_cfg: SceneEntityCfg = _DEFAULT_ASSET_CFG) -> torch.Tensor

Penalize asset pos from its target pos using tanh kernel.

yaw_error_l2

yaw_error_l2(env: ManagerBasedRlEnv, command_name: str, asset_cfg: SceneEntityCfg = _DEFAULT_ASSET_CFG) -> torch.Tensor

Penalize heading error from target heading using L2 squared kernel.

yaw_error_tanh

yaw_error_tanh(env: ManagerBasedRlEnv, command_name: str, std: float = 0.8, asset_cfg: SceneEntityCfg = _DEFAULT_ASSET_CFG) -> torch.Tensor

Penalize heading error from target heading using tanh kernel.

actions

Classes:

Name	Description
ControlAction	Body torque control action term.
ControlActionCfg	See :class:ControlAction for more details.

ControlAction

ControlAction(cfg: ControlActionCfg, env: ManagerBasedRlEnv)

Bases: ActionTerm

Body torque control action term.

This action term applies a wrench to the drone body frame based on action commands

ControlActionCfg dataclass

ControlActionCfg(*, class_type: type[ActionTerm] = ControlAction, entity_name: str = 'robot', params: VehicleParams = VehicleParams())

Bases: ActionTermCfg

See :class:ControlAction for more details.

Attributes:

Name	Type	Description
class_type	type[ActionTerm]	Class of the action term.
entity_name	str	Name of the asset in the environment for which the commands are generated.
params	VehicleParams	Vehicle parameters.

class_type class-attribute instance-attribute

class_type: type[ActionTerm] = ControlAction

Class of the action term.

entity_name class-attribute instance-attribute

entity_name: str = 'robot'

Name of the asset in the environment for which the commands are generated.

params class-attribute instance-attribute

params: VehicleParams = field(default_factory=VehicleParams)

Vehicle parameters.

commands

Classes:

Name	Description
UniformPoseCommandGlobal	Command generator that generates pose commands containing a 3-D position and heading.
UniformPoseCommandGlobalCfg	Configuration for the uniform 3D-pose command generator.

UniformPoseCommandGlobal

UniformPoseCommandGlobal(cfg: UniformPoseCommandGlobalCfg, env: ManagerBasedRlEnv)

Bases: CommandTerm

Command generator that generates pose commands containing a 3-D position and heading. The command generator samples uniform 3D positions around the environment origin. The heading command is either set to point towards the target or is sampled uniformly. This can be configured through the :attr:UniformPoseCommandGlobalCfg.simple_heading parameter in the configuration.

Initialize the command generator class.

Parameters:

Name	Type	Description	Default
cfg	UniformPoseCommandGlobalCfg	The configuration parameters for the command generator.	required
env	ManagerBasedRlEnv	The environment object.	required

Attributes:

Name	Type	Description
cfg	UniformPoseCommandGlobalCfg	Configuration for the command generator.
command	Tensor	The desired 3D-pose in base frame. Shape is (num_envs, 4).

cfg instance-attribute

cfg: UniformPoseCommandGlobalCfg

Configuration for the command generator.

command property

command: Tensor

The desired 3D-pose in base frame. Shape is (num_envs, 4).

UniformPoseCommandGlobalCfg dataclass

UniformPoseCommandGlobalCfg(*, class_type: type = UniformPoseCommandGlobal, entity_name: str, simple_heading: bool, ranges: Ranges, viz: VizCfg = VizCfg())

Bases: CommandTermCfg

Configuration for the uniform 3D-pose command generator.

Classes:

Name	Description
Ranges	Uniform distribution ranges for the position commands.

Attributes:

Name	Type	Description
entity_name	str	Name of the asset in the environment for which the commands are generated.
simple_heading	bool	Whether to use simple heading or not.
ranges	Ranges	Distribution ranges for the position commands.

entity_name instance-attribute

entity_name: str

Name of the asset in the environment for which the commands are generated.

simple_heading instance-attribute

simple_heading: bool

Whether to use simple heading or not.

If True, the heading is in the direction of the target position.

ranges instance-attribute

ranges: Ranges

Distribution ranges for the position commands.

Ranges dataclass

Ranges(pos_x: tuple[float, float], pos_y: tuple[float, float], pos_z: tuple[float, float], heading: tuple[float, float])

Uniform distribution ranges for the position commands.

Attributes:

Name	Type	Description
pos_x	tuple[float, float]	Range for the x position (in m).
pos_y	tuple[float, float]	Range for the y position (in m).
pos_z	tuple[float, float]	Range for the z position (in m).
heading	tuple[float, float]	Heading range for the position commands (in rad).

pos_x instance-attribute

pos_x: tuple[float, float]

Range for the x position (in m).

pos_y instance-attribute

pos_y: tuple[float, float]

Range for the y position (in m).

pos_z instance-attribute

pos_z: tuple[float, float]

Range for the z position (in m).

heading instance-attribute

heading: tuple[float, float]

Heading range for the position commands (in rad).

Used only if :attr:simple_heading is False.

rewards

Functions:

Name	Description
lin_vel_l2	Penalize base linear velocity using L2 squared kernel.
ang_vel_l2	Penalize base angular velocity using L2 squared kernel.
pos_error_l2	Penalize asset pos from its target pos using L2 squared kernel.
pos_error_tanh	Penalize asset pos from its target pos using tanh kernel.
yaw_error_l2	Penalize heading error from target heading using L2 squared kernel.
yaw_error_tanh	Penalize heading error from target heading using tanh kernel.

lin_vel_l2

lin_vel_l2(env: ManagerBasedRlEnv, asset_cfg: SceneEntityCfg = _DEFAULT_ASSET_CFG) -> torch.Tensor

Penalize base linear velocity using L2 squared kernel.

ang_vel_l2

ang_vel_l2(env: ManagerBasedRlEnv, asset_cfg: SceneEntityCfg = _DEFAULT_ASSET_CFG) -> torch.Tensor

Penalize base angular velocity using L2 squared kernel.

pos_error_l2

pos_error_l2(env: ManagerBasedRlEnv, command_name: str, asset_cfg: SceneEntityCfg = _DEFAULT_ASSET_CFG) -> torch.Tensor

Penalize asset pos from its target pos using L2 squared kernel.

pos_error_tanh

pos_error_tanh(env: ManagerBasedRlEnv, command_name: str, std: float = 0.8, asset_cfg: SceneEntityCfg = _DEFAULT_ASSET_CFG) -> torch.Tensor

Penalize asset pos from its target pos using tanh kernel.

yaw_error_l2

yaw_error_l2(env: ManagerBasedRlEnv, command_name: str, asset_cfg: SceneEntityCfg = _DEFAULT_ASSET_CFG) -> torch.Tensor

Penalize heading error from target heading using L2 squared kernel.

yaw_error_tanh

yaw_error_tanh(env: ManagerBasedRlEnv, command_name: str, std: float = 0.8, asset_cfg: SceneEntityCfg = _DEFAULT_ASSET_CFG) -> torch.Tensor

Penalize heading error from target heading using tanh kernel.

trajectories

Trajectory implementations package for LAV2.

Modules:

Name	Description
base	NumPy trajectory base classes.
helix	NumPy helix trajectory implementation.
hover_step	NumPy piecewise hover-step trajectory implementation.
lemniscate	NumPy lemniscate trajectory implementation.
lissajous	NumPy Lissajous trajectory implementation.
rectangle	NumPy rectangle trajectory implementation.
torch	Torch trajectory implementations for batched RL environments.

Classes:

Name	Description
BaseTrajectory	Abstract base class for time-parameterized NumPy reference trajectories.
SpatialTrajectory	Base class for trajectories centered around a world-frame anchor point.
HelixTrajectory	Helix trajectory with optional vertical oscillation.
HoverStepTrajectory	Piecewise-constant position and yaw trajectory.
LemniscateTrajectory	Bernoulli lemniscate trajectory with optional vertical oscillation.
LissajousTrajectory	Three-dimensional Lissajous trajectory.
RectangleTrajectory	Constant-speed rectangle trajectory in the horizontal plane.

BaseTrajectory dataclass

BaseTrajectory(yaw_from_velocity: bool = True, fixed_yaw: float = 0.0, attitude: tuple[float, float] = (0.0, 0.0), body_rate: tuple[float, float] = (0.0, 0.0))

Bases: ABC

Abstract base class for time-parameterized NumPy reference trajectories.

Methods:

Name	Description
sample_heading	Sample yaw and yaw-rate references for time t and translation state.
build_target	Assemble a full 12D target from position, velocity, and acceleration.
sample	Sample the full 12D target at time t.
sample_batch	Sample a batch of timestamps and stack the resulting 12D targets.

sample_heading

sample_heading(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> tuple[float, float]

Sample yaw and yaw-rate references for time t and translation state.

build_target

build_target(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> np.ndarray

Assemble a full 12D target from position, velocity, and acceleration.

sample abstractmethod

sample(t: float) -> np.ndarray

Sample the full 12D target at time t.

sample_batch

sample_batch(ts: ndarray) -> np.ndarray

Sample a batch of timestamps and stack the resulting 12D targets.

SpatialTrajectory dataclass

SpatialTrajectory(yaw_from_velocity: bool = True, fixed_yaw: float = 0.0, attitude: tuple[float, float] = (0.0, 0.0), body_rate: tuple[float, float] = (0.0, 0.0), center: tuple[float, float, float] = (0.0, 0.0, 1.0))

Bases: BaseTrajectory

Base class for trajectories centered around a world-frame anchor point.

Methods:

Name	Description
sample_heading	Sample yaw and yaw-rate references for time t and translation state.
build_target	Assemble a full 12D target from position, velocity, and acceleration.
sample	Sample the full 12D target at time t.
sample_batch	Sample a batch of timestamps and stack the resulting 12D targets.
__post_init__	Cache the trajectory center as a float32 NumPy vector.

sample_heading

sample_heading(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> tuple[float, float]

Sample yaw and yaw-rate references for time t and translation state.

build_target

build_target(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> np.ndarray

Assemble a full 12D target from position, velocity, and acceleration.

sample abstractmethod

sample(t: float) -> np.ndarray

Sample the full 12D target at time t.

sample_batch

sample_batch(ts: ndarray) -> np.ndarray

Sample a batch of timestamps and stack the resulting 12D targets.

__post_init__

__post_init__()

Cache the trajectory center as a float32 NumPy vector.

HelixTrajectory dataclass

HelixTrajectory(yaw_from_velocity: bool = True, fixed_yaw: float = 0.0, attitude: tuple[float, float] = (0.0, 0.0), body_rate: tuple[float, float] = (0.0, 0.0), center: tuple[float, float, float] = (0.0, 0.0, 1.0), radius: float = 1.0, omega: float = 0.5, z_amplitude: float = 0.0, z_frequency: float = 1.0, z_phase: float = 0.0)

Bases: SpatialTrajectory

Helix trajectory with optional vertical oscillation.

Methods:

Name	Description
sample_heading	Sample yaw and yaw-rate references for time t and translation state.
build_target	Assemble a full 12D target from position, velocity, and acceleration.
sample_batch	Sample a batch of timestamps and stack the resulting 12D targets.
__post_init__	Cache the trajectory center as a float32 NumPy vector.
sample	Sample the helix trajectory at time t.

sample_heading

sample_heading(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> tuple[float, float]

Sample yaw and yaw-rate references for time t and translation state.

build_target

build_target(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> np.ndarray

Assemble a full 12D target from position, velocity, and acceleration.

sample_batch

sample_batch(ts: ndarray) -> np.ndarray

Sample a batch of timestamps and stack the resulting 12D targets.

__post_init__

__post_init__()

Cache the trajectory center as a float32 NumPy vector.

sample

sample(t: float) -> np.ndarray

Sample the helix trajectory at time t.

HoverStepTrajectory dataclass

HoverStepTrajectory(yaw_from_velocity: bool = False, fixed_yaw: float = 0.0, attitude: tuple[float, float] = (0.0, 0.0), body_rate: tuple[float, float] = (0.0, 0.0), initial: tuple[float, float, float, float] = (0.0, 0.0, 1.0, 0.0), steps: tuple[tuple[float, tuple[float, float, float, float]], ...] = ((5.0, (1.0, 0.0, 1.0, 0.0)), (10.0, (1.0, 1.0, 1.5, np.pi / 2.0)), (15.0, (0.0, 1.0, 1.5, np.pi))))

Bases: BaseTrajectory

Piecewise-constant position and yaw trajectory.

Methods:

Name	Description
sample_heading	Sample yaw and yaw-rate references for time t and translation state.
build_target	Assemble a full 12D target from position, velocity, and acceleration.
sample_batch	Sample a batch of timestamps and stack the resulting 12D targets.
sample	Sample the active hover target at time t.

sample_heading

sample_heading(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> tuple[float, float]

Sample yaw and yaw-rate references for time t and translation state.

build_target

build_target(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> np.ndarray

Assemble a full 12D target from position, velocity, and acceleration.

sample_batch

sample_batch(ts: ndarray) -> np.ndarray

Sample a batch of timestamps and stack the resulting 12D targets.

sample

sample(t: float) -> np.ndarray

Sample the active hover target at time t.

LemniscateTrajectory dataclass

LemniscateTrajectory(yaw_from_velocity: bool = True, fixed_yaw: float = 0.0, attitude: tuple[float, float] = (0.0, 0.0), body_rate: tuple[float, float] = (0.0, 0.0), center: tuple[float, float, float] = (0.0, 0.0, 1.0), scale: float = 1.0, omega: float = 0.5, z_amplitude: float = 0.0, z_frequency: float = 1.0, z_phase: float = 0.0)

Bases: SpatialTrajectory

Bernoulli lemniscate trajectory with optional vertical oscillation.

Methods:

Name	Description
sample_heading	Sample yaw and yaw-rate references for time t and translation state.
build_target	Assemble a full 12D target from position, velocity, and acceleration.
sample_batch	Sample a batch of timestamps and stack the resulting 12D targets.
__post_init__	Cache the trajectory center as a float32 NumPy vector.
sample	Sample the lemniscate trajectory at time t.

sample_heading

sample_heading(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> tuple[float, float]

Sample yaw and yaw-rate references for time t and translation state.

build_target

build_target(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> np.ndarray

Assemble a full 12D target from position, velocity, and acceleration.

sample_batch

sample_batch(ts: ndarray) -> np.ndarray

Sample a batch of timestamps and stack the resulting 12D targets.

__post_init__

__post_init__()

Cache the trajectory center as a float32 NumPy vector.

sample

sample(t: float) -> np.ndarray

Sample the lemniscate trajectory at time t.

LissajousTrajectory dataclass

LissajousTrajectory(yaw_from_velocity: bool = True, fixed_yaw: float = 0.0, attitude: tuple[float, float] = (0.0, 0.0), body_rate: tuple[float, float] = (0.0, 0.0), center: tuple[float, float, float] = (0.0, 0.0, 1.0), amplitude: tuple[float, float, float] = (1.0, 1.0, 0.2), frequency: tuple[float, float, float] = (1.0, 2.0, 0.5), phase: tuple[float, float, float] = (0.0, np.pi / 2.0, 0.0), omega: float = 0.5)

Bases: SpatialTrajectory

Three-dimensional Lissajous trajectory.

Methods:

Name	Description
sample_heading	Sample yaw and yaw-rate references for time t and translation state.
build_target	Assemble a full 12D target from position, velocity, and acceleration.
sample_batch	Sample a batch of timestamps and stack the resulting 12D targets.
__post_init__	Cache vector-valued parameters as float32 NumPy arrays.
sample	Sample the Lissajous trajectory at time t.

sample_heading

sample_heading(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> tuple[float, float]

Sample yaw and yaw-rate references for time t and translation state.

build_target

build_target(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> np.ndarray

Assemble a full 12D target from position, velocity, and acceleration.

sample_batch

sample_batch(ts: ndarray) -> np.ndarray

Sample a batch of timestamps and stack the resulting 12D targets.

__post_init__

__post_init__()

Cache vector-valued parameters as float32 NumPy arrays.

sample

sample(t: float) -> np.ndarray

Sample the Lissajous trajectory at time t.

RectangleTrajectory dataclass

RectangleTrajectory(yaw_from_velocity: bool = True, fixed_yaw: float = 0.0, attitude: tuple[float, float] = (0.0, 0.0), body_rate: tuple[float, float] = (0.0, 0.0), center: tuple[float, float, float] = (0.0, 0.0, 1.0), size: tuple[float, float] = (2.0, 1.0), speed: float = 0.5)

Bases: SpatialTrajectory

Constant-speed rectangle trajectory in the horizontal plane.

Methods:

Name	Description
sample_heading	Sample yaw and yaw-rate references for time t and translation state.
build_target	Assemble a full 12D target from position, velocity, and acceleration.
sample_batch	Sample a batch of timestamps and stack the resulting 12D targets.
__post_init__	Precompute rectangle corners and edge lengths.
sample	Sample the rectangle trajectory at time t.

sample_heading

sample_heading(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> tuple[float, float]

Sample yaw and yaw-rate references for time t and translation state.

build_target

build_target(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> np.ndarray

Assemble a full 12D target from position, velocity, and acceleration.

sample_batch

sample_batch(ts: ndarray) -> np.ndarray

Sample a batch of timestamps and stack the resulting 12D targets.

__post_init__

__post_init__()

Precompute rectangle corners and edge lengths.

sample

sample(t: float) -> np.ndarray

Sample the rectangle trajectory at time t.

base

NumPy trajectory base classes.

Classes:

Name	Description
BaseTrajectory	Abstract base class for time-parameterized NumPy reference trajectories.
SpatialTrajectory	Base class for trajectories centered around a world-frame anchor point.

BaseTrajectory dataclass

BaseTrajectory(yaw_from_velocity: bool = True, fixed_yaw: float = 0.0, attitude: tuple[float, float] = (0.0, 0.0), body_rate: tuple[float, float] = (0.0, 0.0))

Bases: ABC

Abstract base class for time-parameterized NumPy reference trajectories.

Methods:

Name	Description
sample_heading	Sample yaw and yaw-rate references for time t and translation state.
build_target	Assemble a full 12D target from position, velocity, and acceleration.
sample	Sample the full 12D target at time t.
sample_batch	Sample a batch of timestamps and stack the resulting 12D targets.

sample_heading

sample_heading(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> tuple[float, float]

Sample yaw and yaw-rate references for time t and translation state.

build_target

build_target(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> np.ndarray

Assemble a full 12D target from position, velocity, and acceleration.

sample abstractmethod

sample(t: float) -> np.ndarray

Sample the full 12D target at time t.

sample_batch

sample_batch(ts: ndarray) -> np.ndarray

Sample a batch of timestamps and stack the resulting 12D targets.

SpatialTrajectory dataclass

SpatialTrajectory(yaw_from_velocity: bool = True, fixed_yaw: float = 0.0, attitude: tuple[float, float] = (0.0, 0.0), body_rate: tuple[float, float] = (0.0, 0.0), center: tuple[float, float, float] = (0.0, 0.0, 1.0))

Bases: BaseTrajectory

Base class for trajectories centered around a world-frame anchor point.

Methods:

Name	Description
__post_init__	Cache the trajectory center as a float32 NumPy vector.
sample_heading	Sample yaw and yaw-rate references for time t and translation state.
build_target	Assemble a full 12D target from position, velocity, and acceleration.
sample	Sample the full 12D target at time t.
sample_batch	Sample a batch of timestamps and stack the resulting 12D targets.

__post_init__

__post_init__()

Cache the trajectory center as a float32 NumPy vector.

sample_heading

sample_heading(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> tuple[float, float]

Sample yaw and yaw-rate references for time t and translation state.

build_target

build_target(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> np.ndarray

Assemble a full 12D target from position, velocity, and acceleration.

sample abstractmethod

sample(t: float) -> np.ndarray

Sample the full 12D target at time t.

sample_batch

sample_batch(ts: ndarray) -> np.ndarray

Sample a batch of timestamps and stack the resulting 12D targets.

helix

NumPy helix trajectory implementation.

Classes:

Name	Description
HelixTrajectory	Helix trajectory with optional vertical oscillation.

HelixTrajectory dataclass

HelixTrajectory(yaw_from_velocity: bool = True, fixed_yaw: float = 0.0, attitude: tuple[float, float] = (0.0, 0.0), body_rate: tuple[float, float] = (0.0, 0.0), center: tuple[float, float, float] = (0.0, 0.0, 1.0), radius: float = 1.0, omega: float = 0.5, z_amplitude: float = 0.0, z_frequency: float = 1.0, z_phase: float = 0.0)

Bases: SpatialTrajectory

Helix trajectory with optional vertical oscillation.

Methods:

Name	Description
sample	Sample the helix trajectory at time t.
sample_heading	Sample yaw and yaw-rate references for time t and translation state.
build_target	Assemble a full 12D target from position, velocity, and acceleration.
sample_batch	Sample a batch of timestamps and stack the resulting 12D targets.
__post_init__	Cache the trajectory center as a float32 NumPy vector.

sample

sample(t: float) -> np.ndarray

Sample the helix trajectory at time t.

sample_heading

sample_heading(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> tuple[float, float]

Sample yaw and yaw-rate references for time t and translation state.

build_target

build_target(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> np.ndarray

Assemble a full 12D target from position, velocity, and acceleration.

sample_batch

sample_batch(ts: ndarray) -> np.ndarray

Sample a batch of timestamps and stack the resulting 12D targets.

__post_init__

__post_init__()

Cache the trajectory center as a float32 NumPy vector.

hover_step

NumPy piecewise hover-step trajectory implementation.

Classes:

Name	Description
HoverStepTrajectory	Piecewise-constant position and yaw trajectory.

HoverStepTrajectory dataclass

HoverStepTrajectory(yaw_from_velocity: bool = False, fixed_yaw: float = 0.0, attitude: tuple[float, float] = (0.0, 0.0), body_rate: tuple[float, float] = (0.0, 0.0), initial: tuple[float, float, float, float] = (0.0, 0.0, 1.0, 0.0), steps: tuple[tuple[float, tuple[float, float, float, float]], ...] = ((5.0, (1.0, 0.0, 1.0, 0.0)), (10.0, (1.0, 1.0, 1.5, np.pi / 2.0)), (15.0, (0.0, 1.0, 1.5, np.pi))))

Bases: BaseTrajectory

Piecewise-constant position and yaw trajectory.

Methods:

Name	Description
sample	Sample the active hover target at time t.
sample_heading	Sample yaw and yaw-rate references for time t and translation state.
build_target	Assemble a full 12D target from position, velocity, and acceleration.
sample_batch	Sample a batch of timestamps and stack the resulting 12D targets.

sample

sample(t: float) -> np.ndarray

Sample the active hover target at time t.

sample_heading

sample_heading(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> tuple[float, float]

Sample yaw and yaw-rate references for time t and translation state.

build_target

build_target(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> np.ndarray

Assemble a full 12D target from position, velocity, and acceleration.

sample_batch

sample_batch(ts: ndarray) -> np.ndarray

Sample a batch of timestamps and stack the resulting 12D targets.

lemniscate

NumPy lemniscate trajectory implementation.

Classes:

Name	Description
LemniscateTrajectory	Bernoulli lemniscate trajectory with optional vertical oscillation.

LemniscateTrajectory dataclass

LemniscateTrajectory(yaw_from_velocity: bool = True, fixed_yaw: float = 0.0, attitude: tuple[float, float] = (0.0, 0.0), body_rate: tuple[float, float] = (0.0, 0.0), center: tuple[float, float, float] = (0.0, 0.0, 1.0), scale: float = 1.0, omega: float = 0.5, z_amplitude: float = 0.0, z_frequency: float = 1.0, z_phase: float = 0.0)

Bases: SpatialTrajectory

Bernoulli lemniscate trajectory with optional vertical oscillation.

Methods:

Name	Description
sample	Sample the lemniscate trajectory at time t.
sample_heading	Sample yaw and yaw-rate references for time t and translation state.
build_target	Assemble a full 12D target from position, velocity, and acceleration.
sample_batch	Sample a batch of timestamps and stack the resulting 12D targets.
__post_init__	Cache the trajectory center as a float32 NumPy vector.

sample

sample(t: float) -> np.ndarray

Sample the lemniscate trajectory at time t.

sample_heading

sample_heading(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> tuple[float, float]

Sample yaw and yaw-rate references for time t and translation state.

build_target

build_target(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> np.ndarray

Assemble a full 12D target from position, velocity, and acceleration.

sample_batch

sample_batch(ts: ndarray) -> np.ndarray

Sample a batch of timestamps and stack the resulting 12D targets.

__post_init__

__post_init__()

Cache the trajectory center as a float32 NumPy vector.

lissajous

NumPy Lissajous trajectory implementation.

Classes:

Name	Description
LissajousTrajectory	Three-dimensional Lissajous trajectory.

LissajousTrajectory dataclass

LissajousTrajectory(yaw_from_velocity: bool = True, fixed_yaw: float = 0.0, attitude: tuple[float, float] = (0.0, 0.0), body_rate: tuple[float, float] = (0.0, 0.0), center: tuple[float, float, float] = (0.0, 0.0, 1.0), amplitude: tuple[float, float, float] = (1.0, 1.0, 0.2), frequency: tuple[float, float, float] = (1.0, 2.0, 0.5), phase: tuple[float, float, float] = (0.0, np.pi / 2.0, 0.0), omega: float = 0.5)

Bases: SpatialTrajectory

Three-dimensional Lissajous trajectory.

Methods:

Name	Description
__post_init__	Cache vector-valued parameters as float32 NumPy arrays.
sample	Sample the Lissajous trajectory at time t.
sample_heading	Sample yaw and yaw-rate references for time t and translation state.
build_target	Assemble a full 12D target from position, velocity, and acceleration.
sample_batch	Sample a batch of timestamps and stack the resulting 12D targets.

__post_init__

__post_init__()

Cache vector-valued parameters as float32 NumPy arrays.

sample

sample(t: float) -> np.ndarray

Sample the Lissajous trajectory at time t.

sample_heading

sample_heading(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> tuple[float, float]

Sample yaw and yaw-rate references for time t and translation state.

build_target

build_target(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> np.ndarray

Assemble a full 12D target from position, velocity, and acceleration.

sample_batch

sample_batch(ts: ndarray) -> np.ndarray

Sample a batch of timestamps and stack the resulting 12D targets.

rectangle

NumPy rectangle trajectory implementation.

Classes:

Name	Description
RectangleTrajectory	Constant-speed rectangle trajectory in the horizontal plane.

RectangleTrajectory dataclass

RectangleTrajectory(yaw_from_velocity: bool = True, fixed_yaw: float = 0.0, attitude: tuple[float, float] = (0.0, 0.0), body_rate: tuple[float, float] = (0.0, 0.0), center: tuple[float, float, float] = (0.0, 0.0, 1.0), size: tuple[float, float] = (2.0, 1.0), speed: float = 0.5)

Bases: SpatialTrajectory

Constant-speed rectangle trajectory in the horizontal plane.

Methods:

Name	Description
__post_init__	Precompute rectangle corners and edge lengths.
sample	Sample the rectangle trajectory at time t.
sample_heading	Sample yaw and yaw-rate references for time t and translation state.
build_target	Assemble a full 12D target from position, velocity, and acceleration.
sample_batch	Sample a batch of timestamps and stack the resulting 12D targets.

__post_init__

__post_init__()

Precompute rectangle corners and edge lengths.

sample

sample(t: float) -> np.ndarray

Sample the rectangle trajectory at time t.

sample_heading

sample_heading(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> tuple[float, float]

Sample yaw and yaw-rate references for time t and translation state.

build_target

build_target(t: float, pos: ndarray, vel: ndarray, acc: ndarray) -> np.ndarray

Assemble a full 12D target from position, velocity, and acceleration.

sample_batch

sample_batch(ts: ndarray) -> np.ndarray

Sample a batch of timestamps and stack the resulting 12D targets.

torch

Torch trajectory implementations for batched RL environments.

Modules:

Name	Description
base	Torch trajectory base classes.
helix	Torch helix trajectory implementation.
hover_step	Torch hover-step trajectory implementation.
lemniscate	Torch lemniscate trajectory implementation.
lissajous	Torch Lissajous trajectory implementation.
rectangle	Torch rectangle trajectory implementation.

Classes:

Name	Description
BaseTrajectory	Abstract base class for batched torch trajectory implementations.
SpatialTrajectory	Torch base class for trajectories centered around a batched world-frame anchor.
HelixTrajectory	Batched helix trajectory with optional vertical oscillation.
HoverStepTrajectory	Batched piecewise-constant position and yaw trajectory.
LemniscateTrajectory	Batched Bernoulli lemniscate trajectory with vertical oscillation.
LissajousTrajectory	Batched three-dimensional Lissajous trajectory.
RectangleTrajectory	Batched constant-speed rectangle trajectory in the horizontal plane.

BaseTrajectory dataclass

BaseTrajectory(*, yaw_from_velocity: bool = True, fixed_yaw: Tensor | None = None, attitude: Tensor | None = None, body_rate: Tensor | None = None)

Bases: ABC

Abstract base class for batched torch trajectory implementations.

Methods:

Name	Description
sample_heading	Sample batched yaw and yaw-rate references for time values t.
build_target	Assemble a batched 12D target tensor from motion primitives.
sample	Sample a batched 12D target tensor for time values t.

sample_heading

sample_heading(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> tuple[torch.Tensor, torch.Tensor]

Sample batched yaw and yaw-rate references for time values t.

build_target

build_target(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> torch.Tensor

Assemble a batched 12D target tensor from motion primitives.

sample abstractmethod

sample(t: Tensor) -> torch.Tensor

Sample a batched 12D target tensor for time values t.

SpatialTrajectory dataclass

SpatialTrajectory(*, yaw_from_velocity: bool = True, fixed_yaw: Tensor | None = None, attitude: Tensor | None = None, body_rate: Tensor | None = None, center: Tensor)

Bases: BaseTrajectory

Torch base class for trajectories centered around a batched world-frame anchor.

Methods:

Name	Description
sample_heading	Sample batched yaw and yaw-rate references for time values t.
build_target	Assemble a batched 12D target tensor from motion primitives.
sample	Sample a batched 12D target tensor for time values t.
__post_init__	Initialize shared tensors from the center batch shape when available.

sample_heading

sample_heading(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> tuple[torch.Tensor, torch.Tensor]

Sample batched yaw and yaw-rate references for time values t.

build_target

build_target(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> torch.Tensor

Assemble a batched 12D target tensor from motion primitives.

sample abstractmethod

sample(t: Tensor) -> torch.Tensor

Sample a batched 12D target tensor for time values t.

__post_init__

__post_init__()

Initialize shared tensors from the center batch shape when available.

HelixTrajectory dataclass

HelixTrajectory(*, yaw_from_velocity: bool = True, fixed_yaw: Tensor | None = None, attitude: Tensor | None = None, body_rate: Tensor | None = None, center: Tensor, radius: Tensor, omega: Tensor, z_amplitude: Tensor, z_frequency: Tensor, z_phase: Tensor)

Bases: SpatialTrajectory

Batched helix trajectory with optional vertical oscillation.

Methods:

Name	Description
sample_heading	Sample batched yaw and yaw-rate references for time values t.
build_target	Assemble a batched 12D target tensor from motion primitives.
__post_init__	Initialize shared tensors from the center batch shape when available.
sample	Sample the batched helix trajectory at time values t.

sample_heading

sample_heading(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> tuple[torch.Tensor, torch.Tensor]

Sample batched yaw and yaw-rate references for time values t.

build_target

build_target(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> torch.Tensor

Assemble a batched 12D target tensor from motion primitives.

__post_init__

__post_init__()

Initialize shared tensors from the center batch shape when available.

sample

sample(t: Tensor) -> torch.Tensor

Sample the batched helix trajectory at time values t.

HoverStepTrajectory dataclass

HoverStepTrajectory(*, yaw_from_velocity: bool = False, fixed_yaw: Tensor | None = None, attitude: Tensor | None = None, body_rate: Tensor | None = None, initial: Tensor, steps: tuple[tuple[float, Tensor], ...] = ())

Bases: BaseTrajectory

Batched piecewise-constant position and yaw trajectory.

Methods:

Name	Description
sample_heading	Sample batched yaw and yaw-rate references for time values t.
build_target	Assemble a batched 12D target tensor from motion primitives.
__post_init__	Initialize shared trajectory tensors from the hover-step batch.
sample	Sample the active hover targets for batched time values t.

sample_heading

sample_heading(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> tuple[torch.Tensor, torch.Tensor]

Sample batched yaw and yaw-rate references for time values t.

build_target

build_target(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> torch.Tensor

Assemble a batched 12D target tensor from motion primitives.

__post_init__

__post_init__()

Initialize shared trajectory tensors from the hover-step batch.

sample

sample(t: Tensor) -> torch.Tensor

Sample the active hover targets for batched time values t.

LemniscateTrajectory dataclass

LemniscateTrajectory(*, yaw_from_velocity: bool = True, fixed_yaw: Tensor | None = None, attitude: Tensor | None = None, body_rate: Tensor | None = None, center: Tensor, scale: Tensor, omega: Tensor, z_amplitude: Tensor, z_frequency: Tensor, z_phase: Tensor)

Bases: SpatialTrajectory

Batched Bernoulli lemniscate trajectory with vertical oscillation.

Methods:

Name	Description
sample_heading	Sample batched yaw and yaw-rate references for time values t.
build_target	Assemble a batched 12D target tensor from motion primitives.
__post_init__	Initialize shared tensors from the center batch shape when available.
sample	Sample the batched lemniscate trajectory at time values t.

sample_heading

sample_heading(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> tuple[torch.Tensor, torch.Tensor]

Sample batched yaw and yaw-rate references for time values t.

build_target

build_target(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> torch.Tensor

Assemble a batched 12D target tensor from motion primitives.

__post_init__

__post_init__()

Initialize shared tensors from the center batch shape when available.

sample

sample(t: Tensor) -> torch.Tensor

Sample the batched lemniscate trajectory at time values t.

LissajousTrajectory dataclass

LissajousTrajectory(*, yaw_from_velocity: bool = True, fixed_yaw: Tensor | None = None, attitude: Tensor | None = None, body_rate: Tensor | None = None, center: Tensor, amplitude: Tensor, frequency: Tensor, phase: Tensor, omega: Tensor)

Bases: SpatialTrajectory

Batched three-dimensional Lissajous trajectory.

Methods:

Name	Description
sample_heading	Sample batched yaw and yaw-rate references for time values t.
build_target	Assemble a batched 12D target tensor from motion primitives.
__post_init__	Initialize shared tensors from the center batch shape when available.
sample	Sample the batched Lissajous trajectory at time values t.

sample_heading

sample_heading(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> tuple[torch.Tensor, torch.Tensor]

Sample batched yaw and yaw-rate references for time values t.

build_target

build_target(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> torch.Tensor

Assemble a batched 12D target tensor from motion primitives.

__post_init__

__post_init__()

Initialize shared tensors from the center batch shape when available.

sample

sample(t: Tensor) -> torch.Tensor

Sample the batched Lissajous trajectory at time values t.

RectangleTrajectory dataclass

RectangleTrajectory(*, yaw_from_velocity: bool = True, fixed_yaw: Tensor | None = None, attitude: Tensor | None = None, body_rate: Tensor | None = None, center: Tensor, size: Tensor, speed: Tensor)

Bases: SpatialTrajectory

Batched constant-speed rectangle trajectory in the horizontal plane.

Methods:

Name	Description
sample_heading	Sample batched yaw and yaw-rate references for time values t.
build_target	Assemble a batched 12D target tensor from motion primitives.
__post_init__	Initialize shared tensors from the center batch shape when available.
sample	Sample the batched rectangle trajectory at time values t.

sample_heading

sample_heading(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> tuple[torch.Tensor, torch.Tensor]

Sample batched yaw and yaw-rate references for time values t.

build_target

build_target(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> torch.Tensor

Assemble a batched 12D target tensor from motion primitives.

__post_init__

__post_init__()

Initialize shared tensors from the center batch shape when available.

sample

sample(t: Tensor) -> torch.Tensor

Sample the batched rectangle trajectory at time values t.

base

Torch trajectory base classes.

Classes:

Name	Description
BaseTrajectory	Abstract base class for batched torch trajectory implementations.
SpatialTrajectory	Torch base class for trajectories centered around a batched world-frame anchor.

BaseTrajectory dataclass

BaseTrajectory(*, yaw_from_velocity: bool = True, fixed_yaw: Tensor | None = None, attitude: Tensor | None = None, body_rate: Tensor | None = None)

Bases: ABC

Abstract base class for batched torch trajectory implementations.

Methods:

Name	Description
sample_heading	Sample batched yaw and yaw-rate references for time values t.
build_target	Assemble a batched 12D target tensor from motion primitives.
sample	Sample a batched 12D target tensor for time values t.

sample_heading

sample_heading(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> tuple[torch.Tensor, torch.Tensor]

Sample batched yaw and yaw-rate references for time values t.

build_target

build_target(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> torch.Tensor

Assemble a batched 12D target tensor from motion primitives.

sample abstractmethod

sample(t: Tensor) -> torch.Tensor

Sample a batched 12D target tensor for time values t.

SpatialTrajectory dataclass

SpatialTrajectory(*, yaw_from_velocity: bool = True, fixed_yaw: Tensor | None = None, attitude: Tensor | None = None, body_rate: Tensor | None = None, center: Tensor)

Bases: BaseTrajectory

Torch base class for trajectories centered around a batched world-frame anchor.

Methods:

Name	Description
__post_init__	Initialize shared tensors from the center batch shape when available.
sample_heading	Sample batched yaw and yaw-rate references for time values t.
build_target	Assemble a batched 12D target tensor from motion primitives.
sample	Sample a batched 12D target tensor for time values t.

__post_init__

__post_init__()

Initialize shared tensors from the center batch shape when available.

sample_heading

sample_heading(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> tuple[torch.Tensor, torch.Tensor]

Sample batched yaw and yaw-rate references for time values t.

build_target

build_target(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> torch.Tensor

Assemble a batched 12D target tensor from motion primitives.

sample abstractmethod

sample(t: Tensor) -> torch.Tensor

Sample a batched 12D target tensor for time values t.

helix

Torch helix trajectory implementation.

Classes:

Name	Description
HelixTrajectory	Batched helix trajectory with optional vertical oscillation.

HelixTrajectory dataclass

HelixTrajectory(*, yaw_from_velocity: bool = True, fixed_yaw: Tensor | None = None, attitude: Tensor | None = None, body_rate: Tensor | None = None, center: Tensor, radius: Tensor, omega: Tensor, z_amplitude: Tensor, z_frequency: Tensor, z_phase: Tensor)

Bases: SpatialTrajectory

Batched helix trajectory with optional vertical oscillation.

Methods:

Name	Description
sample	Sample the batched helix trajectory at time values t.
sample_heading	Sample batched yaw and yaw-rate references for time values t.
build_target	Assemble a batched 12D target tensor from motion primitives.
__post_init__	Initialize shared tensors from the center batch shape when available.

sample

sample(t: Tensor) -> torch.Tensor

Sample the batched helix trajectory at time values t.

sample_heading

sample_heading(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> tuple[torch.Tensor, torch.Tensor]

Sample batched yaw and yaw-rate references for time values t.

build_target

build_target(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> torch.Tensor

Assemble a batched 12D target tensor from motion primitives.

__post_init__

__post_init__()

Initialize shared tensors from the center batch shape when available.

hover_step

Torch hover-step trajectory implementation.

Classes:

Name	Description
HoverStepTrajectory	Batched piecewise-constant position and yaw trajectory.

HoverStepTrajectory dataclass

HoverStepTrajectory(*, yaw_from_velocity: bool = False, fixed_yaw: Tensor | None = None, attitude: Tensor | None = None, body_rate: Tensor | None = None, initial: Tensor, steps: tuple[tuple[float, Tensor], ...] = ())

Bases: BaseTrajectory

Batched piecewise-constant position and yaw trajectory.

Methods:

Name	Description
__post_init__	Initialize shared trajectory tensors from the hover-step batch.
sample	Sample the active hover targets for batched time values t.
sample_heading	Sample batched yaw and yaw-rate references for time values t.
build_target	Assemble a batched 12D target tensor from motion primitives.

__post_init__

__post_init__()

Initialize shared trajectory tensors from the hover-step batch.

sample

sample(t: Tensor) -> torch.Tensor

Sample the active hover targets for batched time values t.

sample_heading

sample_heading(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> tuple[torch.Tensor, torch.Tensor]

Sample batched yaw and yaw-rate references for time values t.

build_target

build_target(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> torch.Tensor

Assemble a batched 12D target tensor from motion primitives.

lemniscate

Torch lemniscate trajectory implementation.

Classes:

Name	Description
LemniscateTrajectory	Batched Bernoulli lemniscate trajectory with vertical oscillation.

LemniscateTrajectory dataclass

LemniscateTrajectory(*, yaw_from_velocity: bool = True, fixed_yaw: Tensor | None = None, attitude: Tensor | None = None, body_rate: Tensor | None = None, center: Tensor, scale: Tensor, omega: Tensor, z_amplitude: Tensor, z_frequency: Tensor, z_phase: Tensor)

Bases: SpatialTrajectory

Batched Bernoulli lemniscate trajectory with vertical oscillation.

Methods:

Name	Description
sample	Sample the batched lemniscate trajectory at time values t.
sample_heading	Sample batched yaw and yaw-rate references for time values t.
build_target	Assemble a batched 12D target tensor from motion primitives.
__post_init__	Initialize shared tensors from the center batch shape when available.

sample

sample(t: Tensor) -> torch.Tensor

Sample the batched lemniscate trajectory at time values t.

sample_heading

sample_heading(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> tuple[torch.Tensor, torch.Tensor]

Sample batched yaw and yaw-rate references for time values t.

build_target

build_target(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> torch.Tensor

Assemble a batched 12D target tensor from motion primitives.

__post_init__

__post_init__()

Initialize shared tensors from the center batch shape when available.

lissajous

Torch Lissajous trajectory implementation.

Classes:

Name	Description
LissajousTrajectory	Batched three-dimensional Lissajous trajectory.

LissajousTrajectory dataclass

LissajousTrajectory(*, yaw_from_velocity: bool = True, fixed_yaw: Tensor | None = None, attitude: Tensor | None = None, body_rate: Tensor | None = None, center: Tensor, amplitude: Tensor, frequency: Tensor, phase: Tensor, omega: Tensor)

Bases: SpatialTrajectory

Batched three-dimensional Lissajous trajectory.

Methods:

Name	Description
sample	Sample the batched Lissajous trajectory at time values t.
sample_heading	Sample batched yaw and yaw-rate references for time values t.
build_target	Assemble a batched 12D target tensor from motion primitives.
__post_init__	Initialize shared tensors from the center batch shape when available.

sample

sample(t: Tensor) -> torch.Tensor

Sample the batched Lissajous trajectory at time values t.

sample_heading

sample_heading(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> tuple[torch.Tensor, torch.Tensor]

Sample batched yaw and yaw-rate references for time values t.

build_target

build_target(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> torch.Tensor

Assemble a batched 12D target tensor from motion primitives.

__post_init__

__post_init__()

Initialize shared tensors from the center batch shape when available.

rectangle

Torch rectangle trajectory implementation.

Classes:

Name	Description
RectangleTrajectory	Batched constant-speed rectangle trajectory in the horizontal plane.

RectangleTrajectory dataclass

RectangleTrajectory(*, yaw_from_velocity: bool = True, fixed_yaw: Tensor | None = None, attitude: Tensor | None = None, body_rate: Tensor | None = None, center: Tensor, size: Tensor, speed: Tensor)

Bases: SpatialTrajectory

Batched constant-speed rectangle trajectory in the horizontal plane.

Methods:

Name	Description
sample	Sample the batched rectangle trajectory at time values t.
sample_heading	Sample batched yaw and yaw-rate references for time values t.
build_target	Assemble a batched 12D target tensor from motion primitives.
__post_init__	Initialize shared tensors from the center batch shape when available.

sample

sample(t: Tensor) -> torch.Tensor

Sample the batched rectangle trajectory at time values t.

sample_heading

sample_heading(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> tuple[torch.Tensor, torch.Tensor]

Sample batched yaw and yaw-rate references for time values t.

build_target

build_target(t: Tensor, pos: Tensor, vel: Tensor, acc: Tensor) -> torch.Tensor

Assemble a batched 12D target tensor from motion primitives.

__post_init__

__post_init__()

Initialize shared tensors from the center batch shape when available.