unilab.visualization.render_many

MuJoCo-only batched rendering helpers.

This module renders many MuJoCo states into image frames by constructing MuJoCo model/data/renderer objects inside worker processes. It is not available for Motrix-only workflows.

Functions

get_grid_offsets(num_envs[, spacing])

init_worker(model_path, shape)

Initialize MuJoCo-only rendering context for a worker process.

render_frame_job(args)

Worker function to render a single frame. args: (state_batch, offsets, transparent, cam_distance, cam_elevation, cam_azimuth, cam_lookat, marker_positions) marker_positions: optional (num_envs, 3) world-frame positions for overlay spheres.

render_frame_tracking_job(args)

Render a single frame with camera tracking on the primary env's root body.

render_states_get_frames(state_list, model_path)

Render a list of physics states and return the list of frames.

render_states_get_frames_tracking(...[, ...])

Render with camera tracking on a single primary environment.

render_states_to_video(state_list, ...[, ...])

Render a list of physics states to a video file using parallel processing.
unilab.visualization.render_many.get_grid_offsets(num_envs, spacing=1.0)[source]
unilab.visualization.render_many.init_worker(model_path, shape)[source]

Initialize MuJoCo-only rendering context for a worker process.

unilab.visualization.render_many.render_frame_job(args)[source]

Worker function to render a single frame. args: (state_batch, offsets, transparent, cam_distance, cam_elevation, cam_azimuth,

cam_lookat, marker_positions)

marker_positions: optional (num_envs, 3) world-frame positions for overlay spheres.

unilab.visualization.render_many.render_states_get_frames(state_list, model_path, width=1280, height=720, num_processes=8, camera_id=-1, cam_distance=2.0, cam_elevation=-20, cam_azimuth=90, cam_lookat=None, render_spacing=1.0, marker_positions_list=None)[source]

Render a list of physics states and return the list of frames.

Parameters:

  • state_list – List of numpy arrays, each shape (num_envs, state_dim).

  • model_path – Path to the mujoco XML model file.

  • width – Width of the video.

  • height – Height of the video.

  • num_processes – Number of parallel processes to use.

  • camera_id – Camera ID to render from.

  • cam_distance – Camera distance from lookat point.

  • cam_elevation – Camera elevation angle in degrees.

  • cam_azimuth – Camera azimuth angle in degrees.

  • cam_lookat – Optional [x, y, z] lookat override for the free camera.

  • render_spacing – Grid spacing used to offset each env in composed video frames.

  • marker_positions_list – Optional list of (num_envs, 3) arrays for overlay spheres.

Returns:

List of numpy arrays (H, W, 3) (RGB)

unilab.visualization.render_many.render_frame_tracking_job(args)[source]

Render a single frame with camera tracking on the primary env’s root body.

The camera uses mjCAMERA_TRACKING so it follows the robot each frame. Only the primary env + nearest neighbours are rendered.

unilab.visualization.render_many.render_states_get_frames_tracking(state_list, model_path, width=1280, height=720, tracking_env_idx=0, max_extra_envs=2, cam_distance=2.0, cam_elevation=-20, cam_azimuth=90, render_spacing=1.0, marker_positions_list=None)[source]

Render with camera tracking on a single primary environment.

Only the primary env and its nearest neighbours are shown. The camera follows the root body of the primary env each frame (mjCAMERA_TRACKING).

Parameters:

  • state_list – List of numpy arrays, each shape (num_envs, state_dim).

  • model_path – Path to the mujoco XML model file.

  • tracking_env_idx – Index of the primary environment to track.

  • max_extra_envs – Number of nearest-neighbour envs to render alongside.

  • cam_distance – Camera distance from the tracked body.

  • cam_elevation – Camera elevation angle in degrees.

  • cam_azimuth – Camera azimuth angle in degrees.

  • render_spacing – Grid spacing for env layout.

unilab.visualization.render_many.render_states_to_video(state_list, model_path, output_path, fps=30, width=1280, height=720, num_processes=8, cam_distance=2.0, cam_elevation=-20, cam_azimuth=90, cam_lookat=None, render_spacing=1.0)[source]

Render a list of physics states to a video file using parallel processing.