Compare qpos/qvel of two rollouts assumed to share model layout.

Return ``{suite_name: [task_name, ...]}`` from LIBERO's benchmark dict. Uses the authoritative ``benchmark.get_benchmark_dict()`` ordering (so ta

Dispersive Loss.

Download the InteriorAgent dataset from Hugging Face.

Download RoboVerse-curated Kujiale USD scenes from Hugging Face.

Download URDF files for visualization.

Download URDF files for visualization.

Download URDF files for visualization.

Download URDF files for all objects and robots in the scenario.

Download URDF files for all objects and robots in the scenario.

Convert VLA action to end-effector control and solve IK. Args: obs: Observation from environment Returns:

Δ-pose (local) -> target EE pose -> cuRobo IK -> joint targets.

Mjlab ``dr.encoder_bias`` port — per-joint encoder calibration offset. Stored on ``env._encoder_bias`` as a ``(num_envs, num_joints)`` tensor.

Ensure environment is in clean initial state with intelligent validation.

Write the procedurally-built handover trajectory to ``path``. The dataset is deterministic and cheap to build, so it is regenerated every tim

Download any ``file=`` ref missing locally from HF (no-op when a clone exists).

Reproduce MoveNearGoogle reset selection (episode_id drawn on RandomState(seed)).

Run the scripted policy over many seeds; aggregate return + success rate. Proves a *distributional* performance match: the metric a model would b

Evaluate the policy for a specified number of episodes. Args: env: The environment to evaluate on actor: The policy network

Evaluate a trained RSL-RL PPO policy

Evaluate a trained RSL-RL PPO policy

Evaluate a single episode.

Evaluate a single episode and save results.

Evaluate a single episode.

Evaluate and collect successful lift trajectories.

Expand all tensors in nested dictionary or list or tuple at a single dimension @dim by @size. Args: x (dict or list or tuple): a

Expand a tensor at a single dimension @dim by @size Args: x (torch.Tensor): input tensor size (int): size to expand

Pull the actor's MLP Linear layers out as a standalone, BC-loadable dict. Keys are kept *relative to the actor and unchanged* (e.g. ``mlp.0.weigh

Extract ``assets/<rel>`` into the overlay; return True if written/exists.

Find all mesh and texture files referenced in the MJCF file.

Return the first existing local HDRI directory containing .hdr files.

Return the first existing local material .blend file.

Recursively find all .urdf files in directory and its subdirectories.

Flatten all tensors in nested dictionary or list or tuple, from @begin_axis onwards. Args: x (dict or list or tuple): a possibly nes

Flat multiple Box action spaces into a single Box space.

Flatten a nested dict or list to a list. For example, given a dict { a: 1 b: { c: 2 } c: 3

Flatten a tensor in all dimensions from @begin_axis onwards. Args: x (torch.Tensor): tensor to flatten begin_axis (int): whi

Samples N random crops for each input in the batch, and then reshapes inputs to [B * N, ...].

Build a normalized action command.

Convenience method to convert pygame surface local coordinates to pymunk coordinates

Compute gait phase based on episode length buffer.

Compute gait phase based on episode length buffer.

Apply @gather_along_dim_with_dim_single to all tensors in a nested dictionary or list or tuple. Args: x (dict or list or tuple):

Generate joint targets for Franka robot without IK solver. Args: step: Current step number num_steps: Total number of steps

Given a batch of sequences of observations, generate a batch of sequences of latents. Inputs: seq_obses: Batch of sequences

Generate a log file with conversion results.

Mjlab ``dr.geom_friction`` port — startup-mode DR. Mutates ``physics.model.geom_friction[gid, axis] = sample`` for each selected geom and eac

Returns a single action to be directly executed. For action chunking policies it either uses an previsouly predicted action chunk, or if it ha

Get the dimension of the action space. :param action_space: :return:

Get actions for current step across all environments.

Construct arguments to `Vec` according to axis index.

Construct arguments to `Vec` according to axis index.

Get the class wrappers. Example: get_class("ReachOrigin", "Cfg") -> ReachOriginCfg get_class("reach_origin", "Cfg") -> ReachOrigi

Get configuration with command line argument support.

Compute the decay factor for the exponential moving average.

Retrieve PyTorch device. It checks that the requested device is available first. For now, it supports only cpu and cuda. By default,

Get a Sapien.Entity given the name. Args: entities (List[sapien.Entity]): entities (link, joint, ...) to query. name (str): name

Return Euler (roll, pitch, yaw) in [-π, π] for a single quaternion. Input: (4,), Output: (3,)

Get gapartnet annotation

Return gravity vector components from a quaternion (w, x, y, z).