Parse command line arguments.

Parse a command payload into a canonical command object.

Extract link (body) info and associated elements (geoms) from MuJoCo. Returns: l_info (dict), j_info (dict)

Extract link (body) info and associated elements (geoms) from MuJoCo. Returns: l_info (dict), j_info (dict)

Extract link (body) info and associated elements (geoms) from MuJoCo. Returns: l_info (dict), j_info (dict)

Parse material/texture information from MuJoCo geom.

Wraps MuJoCo's XML parsing to return an MjModel.

Wraps MuJoCo's XML parsing to return an MjModel.

Wraps MuJoCo's XML parsing to return an MjModel.

Add a cube (free joint) + a static table to an MJCF for manipulation tasks. Used for lift_cube_yam_v2: yam.xml has just the arm; we inject the cu

Plot the ctrl applied per control step.

Make a 2x2 panel: mean reward, value loss, mean episode length, action std.

obs_dict: str: B,To,* return: B,Ta,Da

obs_dict: must include "obs" key result: must include "action" key

obs_dict: must include "obs" key result: must include "action" key

Predict action from current observation. Returns: torch.Tensor: Action tensor of shape (B, 7) containing [dx, dy, dz, dr

Select previous joint

Select previous joint

Select the previous joint.

Apply mjlab BaseAction processing: ``scale * raw + offset``, clip, subtract encoder_bias. The encoder_bias is subtracted only if present on e

Recursively process a body and its children, building up link/joint dicts.

Recursively process a body and its children, building up link/joint dicts.

Recursively process a body and its children, building up link/joint dicts.

Process common input (TAB, J, C, ESC)

Process common input (TAB, J, C, ESC) - separate function

Process shared keyboard input (TAB, J, C, ESC) for mode toggles.

Recursively process all MJCF/XML files in the given folder and its subfolders. Args: folder_path (str): Path to the folder containin

Process labels according to scene points and object poses.

Process keyboard input - returns delta position, delta rotation, joint angle delta, and control flags

Process keyboard input - returns delta position, delta rotation, joint angle delta, and control flags

Process keyboard input and return motion deltas plus control flags.

Process a single MJCF file and save a snapshot.

Processes the observation to be used by the policy, according to the observation the policy is configured to use.

Process a single URDF file and take a snapshot using Isaac Gym.

Process a single USD file and take a snapshot using Isaac Lab.

Return available transform profile names.

Projection operation that includes q_support directly

Projection operation that includes q_support directly

Assemble the 9-d proprio vector from a LIBERO obs dict (native or demo). joint positions (7) + gripper (2). Native eval obs uses ``robot0_joint_p

Collect configured keys and push them as one entry. The values are taken from env_states.robots[env_name] or from attributes on the H

data: key: T,*

Rotate ``v`` from body frame into world frame using ``q`` (xyzw).

Convert an XYZW quaternion to a 3x3 rotation matrix.

Randomize material properties across supported simulators.

Randomize masses and optionally recompute inertias.

Randomize the query.

Randomize the query.

Apply randomization for a specific variant. Args: variant_id: Variant ID (0, 1, 2, ...)

Register all dm_control suite tasks under DMControl/<domain>-<task> namespace.

Register all gymnasium-robotics envs under GymRobotics/<env_id> namespace.

Register every LIBERO task as ``Libero/<suite>__<task_name>`` (lazy). Idempotent. Returns the registered ids. Safe when LIBERO is absent (returns

Register every LIBERO-plus task as ``LiberoPlus/<suite>__<task>`` (lazy). Idempotent. Returns the registered ids. Safe when LIBERO-plus is absent

Register all ManiSkill 3 envs under a 'ManiSkill3/<env_id>' namespace. Idempotent. Returns list of registered (or already-present) env ids.

Register all Meta-World MT50 train + test tasks under MetaWorld/<name> namespace.

Register all mjlab tasks as gym envs accessible via gym.make(prefix + task_id). Idempotent — gymnasium silently ignores re-registration. Retu

Register all myosuite envs under 'MyoSuite/<env_id>' namespace. Returns list of registered (or already-present) env ids.

Register all known robosuite envs under ``<prefix><EnvName>``. Idempotent.

Register every RoboTwin task as ``RoboTwin/<name>`` (lazy entry point). Idempotent. Returns the registered ids. Safe when RoboTwin/its deps are

Remap a dataset demo's embedded MJCF to the locally-installed robosuite. ``fixer.edit_model_xml`` rebases the collector's absolute paths to the l

Rebase a LIBERO-plus perturbed ``model_file`` to local assets. Same as :func:`remap_libero_model` (robosuite + chiliocosm roots, dm_control-safe)

Policy-facing overhead image (uint8) with greenscreen overlay applied.

Render two qpos sequences (robosuite-side, handler-side) with visual meshes and an objective per-frame success banner. Used for action-replay vide

Render a demo's state-replay on both stacks with an objective per-frame success flag (robosuite's own ``_check_success`` evaluated at each state).

Collect a short rollout and return a list of RGB frames (H, W, 3, uint8). Works with FastTD3EnvWrapper: render_env.render() must retu

predicate: Return true if the module is to be replaced. func: Return new module to use.

Replay trajectory for a single variant. Args: env: Task environment scenario: Scenario configuration all_states: All traj

Replay trajectory for a single variant. Args: env: Task environment scenario: Scenario configuration all_states: All traj