mjlab.viewer

Viewer module for environment visualization.

Classes

ViewerConfig BaseViewer

NativeMujocoViewer ViserPlayViewer

OffscreenRenderer

Protocols

EnvProtocol

PolicyProtocol

VerbosityLevel

ViewerConfig

class mjlab.viewer.ViewerConfig

lookat: tuple[float, float, float] = (0.0, 0.0, 0.0)

distance: float = 5.0

fovy: float | None = None

elevation: float = -45.0

azimuth: float = 90.0

class OriginType

The frame in which the camera position and target are defined.

AUTO = 1

Track the first non-fixed body, or fall back to a free camera.

WORLD = 2

Free camera at the configured lookat point.

ASSET_ROOT = 3

Track the root body of the asset defined by entity_name.

ASSET_BODY = 4

Track the body defined by body_name in the asset defined by entity_name.

origin_type: OriginType = 1

entity_name: str | None = None

body_name: str | None = None

env_idx: int = 0

max_extra_envs: int = 2

Number of neighboring environments to render around env_idx.

enable_reflections: bool = True

enable_shadows: bool = True

height: int = 240

width: int = 320

BaseViewer

class mjlab.viewer.BaseViewer

Bases: ABC

Abstract base class for environment viewers.

__init__(env: EnvProtocol, policy: PolicyProtocol, frame_rate: float = 30.0, verbosity: int = VerbosityLevel.SILENT)

tick() → bool

Advance one tick: drain actions, step physics, maybe render.

Returns True when a render frame was produced, False otherwise.

NativeMujocoViewer

class mjlab.viewer.NativeMujocoViewer

Bases: BaseViewer

__init__(env: EnvProtocol, policy: PolicyProtocol, frame_rate: float = 60.0, key_callback: Callable[[int], None] | None = None, plot_cfg: PlotCfg | None = None, enable_perturbations: bool = True, verbosity: VerbosityLevel = VerbosityLevel.SILENT)

setup() → None

Setup MuJoCo viewer resources.

sync_env_to_viewer() → None

Copy env state to viewer; update reward figures; render other envs.

sync_viewer_to_env() → None

Sync mouse perturbation to sim via qfrc_applied.

Mouse perturbation forces are converted from Cartesian body space (xfrc_applied) to generalized joint space (qfrc_applied) so that they coexist with programmatic forces on xfrc_applied. See the module docstring for details on the channel separation.

close() → None

Close viewer and cleanup.

reset_environment() → None

Extend BaseViewer.reset_environment to clear reward histories.

ViserPlayViewer

class mjlab.viewer.ViserPlayViewer

Bases: BaseViewer

Interactive Viser-based viewer with playback controls.

__init__(env: EnvProtocol, policy: PolicyProtocol, frame_rate: float = 60.0, verbosity: VerbosityLevel = VerbosityLevel.SILENT, viser_server: viser.ViserServer | None = None, checkpoint_manager: CheckpointManager | None = None) → None

setup() → None

Setup the viewer resources.

sync_env_to_viewer() → None

Synchronize environment state to viewer.

sync_viewer_to_env() → None

Synchronize viewer state to environment (e.g., perturbations).

reset_environment() → None

Extend BaseViewer.reset_environment to clear reward and metrics histories.

close() → None

Close the viewer and cleanup resources.

is_running() → bool

Check if viewer is running.

OffscreenRenderer

class mjlab.viewer.OffscreenRenderer

Bases: object

__init__(model: MjModel, cfg: ViewerConfig, scene: Scene) → None

update(data: Any, debug_vis_callback: Callable[[MujocoNativeDebugVisualizer], None] | None = None, camera: str | None = None) → None

Update renderer with simulation data.

Protocols

class mjlab.viewer.EnvProtocol

Interface we expect from RL environments, which can be either vanilla ManagerBasedRlEnv objects or wrapped with VideoRecorder, RslRlVecEnvWrapper, etc.

__init__(*args, **kwargs)

class mjlab.viewer.PolicyProtocol

__init__(*args, **kwargs)

class mjlab.viewer.VerbosityLevel

__new__(value)