rlightning.env.maniskill_env

ManiSkill environment wrapper.

class rlightning.env.maniskill_env.ManiskillEnv(config: Any, worker_index: int | None = None, preprocess_fn: Callable | None = <function default_env_preprocess_fn>, **kwargs: Any)

Bases: BaseEnv

ManiSkill environment wrapper for vision-based tasks.

add_new_frames(infos: Dict[str, Any], rewards: MockModule('torch.Tensor') | None = None) → None

Append a rendered frame to the video buffer.

add_new_frames_from_obs(raw_obs: Dict[str, Any]) → None

For debugging render

apply_evaluate_cfg() → None

Apply evaluate_cfg overrides to environment member variables.

capture_image(infos: Dict[str, Any] | None = None)

Render the environment and optionally overlay info.

clear_memory()

property device: MockModule('torch.device')

Return the device used by the underlying environment.

property elapsed_steps: MockModule('torch.Tensor')

Return the elapsed steps tensor from the environment.

flush_video(video_sub_dir: str | None = None) → None

Write buffered frames to disk and reset the buffer.

get_action_space()

Return the environment action space.

get_observation_space()

Return the environment observation space.

property instruction: Any

Return the current language instruction.

property is_offloaded: bool

Check if the environment is currently offloaded.

offload()

Offload the environment to free GPU memory. Only deletes the environment object, other variables are preserved.

reload()

Reload the environment. Recreates the environment object only, other variables are preserved.

render(info: Any, rew: Any | None = None)

Render a frame with optional info and rewards.

reset(options: Dict[str, Any] | None = None, **kwargs: Any) → EnvRet

Reset the environment and return an EnvRet.

restore_evaluate_cfg() → None

Restore previously overridden member variables from evaluate_cfg.

sample_action_space() → Any

Sample a random action from the action space.

step(policy_resp: PolicyResponse, auto_reset: bool = True) → EnvRet

Step the environment with a policy response.

property total_num_group_envs: int

Return the total number of grouped environments.

update_reset_state_ids() → None

Update reset state IDs for fixed-reset episodes.