Return a directory containing at least one HDRI, downloading if needed. Args: download: if True, fetch from HF when no local copy is foun

Get indexes of items matching the candidates patterns.

Get the indexes of the bodies matching the sub_names.

Given a set of observation representation and generated latents, get the encoded latent and the loss. Inputs: input_action:

Get the path to load the model from.

Get the path to load the model from.

Get the log directory (aligned with ppo.py saving logic).

Get the log directory.

Return the path to ``material_lib_v2.blend``, downloading if needed. The file is ~1.5 GB. Callers that don't truly need it should pass ``down

Extract mesh and texture information from MJCF/XML file. Returns: mesh_files (dict): Mapping of mesh names to their file paths

Extract mesh and texture information from MJCF/XML file. Returns: mesh_files (dict): Mapping of mesh names to their file paths

Extract mesh and texture information from MJCF/XML file. Returns: mesh_files (dict): Mapping of mesh names to their file paths

Gets the next state of the motion.

Get the shape of the observation (useful for the buffers). :param observation_space: :return:

This long function is unfortunately doing something very simple and is being very defensive: We are separating out all parameters of

Common shapes T,D T,N,D T,H,W,C T,N,H,W,C

Default optimizer class. Override this if you want to use a different optimizer.

Get the point cloud from the observation.

name: resnet18, resnet34, resnet50

Get the real height of an asset from its URDF file.

Check if all environments have run out of actions.

Added kwargs vs diffuser's original implementation Unified API to get any scheduler from its name. Args: name (`str` or `Schedu

Get the controller state.

Implements temporal ensembline, as in Aloha ACT. Takes in a current prediction chunk and returns a single ensembled action

Query the terrain height at given (x, y) positions. All simulators now center the terrain at world origin (0,0). The terrain center is at hei

Convert MuJoCo geometry to URDF geometry dictionary.

Convert MuJoCo geometry to URDF geometry dictionary.

Return ``(left_force_vec, right_force_vec)`` (N) between each finger and the object.

Hash a string or tuple of strings into a stable key.

Hash the names.

Local roboverse_data path that mirrors mjlab-internal ``relpath``.

Local roboverse_data path mirroring the RoboTwin-internal ``relpath``.

This function is used to infer the grasp from the point cloud.

Inference grasp from point cloud Args: cloud_masked (np.ndarray): masked point cloud max_grasps (int, optional): max

Initialize a GO command, enabling torque mode on strong motors.

Initialize an HG command with default motor fields.

Sample material buckets and cache body indices for randomization.

Validate configuration and cache default masses.

Initialize the query.

Initialize the query.

Initialize the query.

Initialize all randomizers showcasing the new architecture.

True when the Panda is grasping ``object_name`` (ManiSkill criterion).

IsaacGym-specific LiDAR point cloud computation.

Isaac Sim LiDAR point cloud query using Isaac Lab LidarSensor. Returns a dict containing local and world point clouds for the target robot.

Yield rendered agentview frames by setting the full state every frame.

Closed-loop servo for robosuite ``Lift``: reach → descend → grasp → lift.

Return canonical benchmark task names.

Return list of object names (keys) from object_init_list.json.

Return list of robot names (keys) from robot_init_list.json.

Load checkpoint from file.

Load checkpoint from file.

Load configuration from YAML file.

Load assets from an EmbodiedGen layout and convert them to the specified format.

Load the initial poses of objects and robots from an EmbodiedGen layout.

Accept either JSONL (preferred) or the stdout text format (fallback).

Load the managers for the environment. This function is responsible for creating the various managers (action, observation, events, e

Decode and validate a UTF-8 JSON teleoperation packet.

Load actor weights from an RSL-RL on-policy-runner checkpoint (mjlab cartpole format).

Return {tag: [(step, value), ...]} for all scalar tags.

Load state list from pkl file.

Load state list from pkl file.

Load state list from pkl file.

Load state list from pkl file.

Load state list from pkl file.

Load state list from pkl file.

Rewrite ``file=`` refs in a stored LIBERO model_file to local abspaths. Returns ``(fixed_xml, info)``. ``info`` has ``n_fixed`` and ``missing``.