mjlab.envs

RL environment classes.

Classes:

ManagerBasedRlEnv	Manager-based RL environment.
ManagerBasedRlEnvCfg	Configuration for a manager-based RL environment.

class mjlab.envs.ManagerBasedRlEnv

Bases: object

Manager-based RL environment.

Attributes:

is_vector_env
metadata
cfg
num_envs	Number of parallel environments.
physics_dt	Physics simulation step size.
step_dt	Environment step size (physics_dt * decimation).
device	Device for computation.
max_episode_length_s	Maximum episode length in seconds.
max_episode_length	Maximum episode length in steps.
unwrapped	Get the unwrapped environment (base case for wrapper chains).

Methods:

__init__(cfg, device[, render_mode])
setup_manager_visualizers()
load_managers()	Load and initialize all managers.
reset(*[, seed, env_ids, options])
step(action)
render()
close()
seed([seed])
update_visualizers(visualizer)

is_vector_env = True

metadata = {'mujoco_version': '3.4.1', 'render_modes': [None, 'rgb_array'], 'warp_version': warp.config.version}

__init__(cfg: ManagerBasedRlEnvCfg, device: str, render_mode: str | None = None, **kwargs) → None

cfg: ManagerBasedRlEnvCfg

property num_envs: int

Number of parallel environments.

property physics_dt: float

Physics simulation step size.

property step_dt: float

Environment step size (physics_dt * decimation).

property device: str

Device for computation.

property max_episode_length_s: float

Maximum episode length in seconds.

property max_episode_length: int

Maximum episode length in steps.

property unwrapped: ManagerBasedRlEnv

Get the unwrapped environment (base case for wrapper chains).

setup_manager_visualizers() → None

load_managers() → None

Load and initialize all managers.

Order is important! Event and command managers must be loaded first, then action and observation managers, then other RL managers.

reset(*, seed: int | None = None, env_ids: Tensor | None = None, options: dict[str, Any] | None = None) → tuple[Dict[str, Tensor | Dict[str, Tensor]], dict]

step(action: Tensor) → tuple[Dict[str, Tensor | Dict[str, Tensor]], Tensor, Tensor, Tensor, dict]

render() → ndarray | None

close() → None

static seed(seed: int = -1) → int

update_visualizers(visualizer: DebugVisualizer) → None

class mjlab.envs.ManagerBasedRlEnvCfg

Bases: object

Configuration for a manager-based RL environment.

Attributes:

decimation	Number of simulation steps per environment step.
scene
observations
actions
events
seed
sim
viewer
episode_length_s	Duration of an episode (in seconds).
rewards	Reward terms configuration.
terminations	Termination terms configuration.
commands	Command terms configuration.
curriculum	Curriculum terms configuration.
is_finite_horizon	Whether the task has a finite or infinite horizon.

Methods:

__init__(*, decimation, scene, observations, ...)

decimation: int

Number of simulation steps per environment step.

scene: SceneCfg

observations: dict[str, ObservationGroupCfg]

actions: dict[str, ActionTermCfg]

events: dict[str, EventTermCfg]

seed: int | None = None

sim: SimulationCfg

viewer: ViewerConfig

episode_length_s: float = 0.0

Duration of an episode (in seconds).

Episode length in steps is computed as:

ceil(episode_length_s / (sim.mujoco.timestep * decimation))

rewards: dict[str, RewardTermCfg]

Reward terms configuration.

terminations: dict[str, TerminationTermCfg]

Termination terms configuration.

commands: dict[str, CommandTermCfg] | None = None

Command terms configuration. If None, no commands are used.

curriculum: dict[str, CurriculumTermCfg] | None = None

Curriculum terms configuration. If None, no curriculum is used.

is_finite_horizon: bool = False

Whether the task has a finite or infinite horizon. Defaults to False (infinite).

• Finite horizon (True): The time limit defines the task boundary. When reached, no future value exists beyond it, so the agent receives a terminal done signal.

• Infinite horizon (False): The time limit is an artificial cutoff. The agent receives a truncated done signal to bootstrap the value of continuing beyond the limit.

__init__(*, decimation: int, scene: ~mjlab.scene.scene.SceneCfg, observations: dict[str, ~mjlab.managers.manager_term_config.ObservationGroupCfg], actions: dict[str, ~mjlab.managers.manager_term_config.ActionTermCfg], events: dict[str, ~mjlab.managers.manager_term_config.EventTermCfg] = <factory>, seed: int | None = None, sim: ~mjlab.sim.sim.SimulationCfg = <factory>, viewer: ~mjlab.viewer.viewer_config.ViewerConfig = <factory>, episode_length_s: float = 0.0, rewards: dict[str, ~mjlab.managers.manager_term_config.RewardTermCfg] = <factory>, terminations: dict[str, ~mjlab.managers.manager_term_config.TerminationTermCfg] = <factory>, commands: dict[str, ~mjlab.managers.manager_term_config.CommandTermCfg] | None = None, curriculum: dict[str, ~mjlab.managers.manager_term_config.CurriculumTermCfg] | None = None, is_finite_horizon: bool = False) → None