unisim.backend.mujoco.motion_export

MuJoCo-only forward-kinematics export for motion-tracking NPZ conversion.

Cold-path tooling shared by the scripts/motion/ CSV-to-NPZ converters. It injects track_* sensors into a model, replays a (root + named joints) trajectory through mj_forward, and reads back joint/body states with the same track_*-sensor-first semantics the training backend uses.

Functions

compute_tracking_fk(model_file, *, ...[, ...])

Replay a root+joint trajectory and return tracking FK arrays.
unisim.backend.mujoco.motion_export.compute_tracking_fk(model_file, *, joint_names, base_poss, base_rots, base_lin_vels, base_ang_vels, dof_poss, dof_vels, progress=False, progress_desc=None)[source]

Replay a root+joint trajectory and return tracking FK arrays.

Parameters:

  • model_file (str) – MuJoCo XML scene file. track_* sensors are injected on a temporary copy; the source file is not modified.

  • joint_names (Sequence[str]) – Actuated joint names in the desired output column order.

  • base_poss (ndarray) – (N, 3) root positions, world frame.

  • base_rots (ndarray) – (N, 4) root quaternions, wxyz.

  • base_lin_vels (ndarray) – (N, 3) root linear velocities, world frame.

  • base_ang_vels (ndarray) – (N, 3) root angular velocities, world frame.

  • dof_poss (ndarray) – (N, len(joint_names)) joint positions.

  • dof_vels (ndarray) – (N, len(joint_names)) joint velocities.

  • progress (bool) – Show a tqdm progress bar over frames.

  • progress_desc (str | None) – Optional tqdm description.

Return type:

dict[str, ndarray]

Returns:

Dict with float32 arrays joint_pos / joint_vel (N, len(joint_names)) and body_pos_w / body_quat_w / body_lin_vel_w / body_ang_vel_w (N, nbody, …) in MuJoCo body-id layout, including the implicit world body 0.