Example: Flatten TensorState into 1D observation tensor for RL algorithms. This function demonstrates how to extract and concatenate various stat

Build a joint target command.

This function indexes out a target dimension of a tensor in a structured way, by allowing a different value to be selected for each member of

View sampling on a unit sphere using Fibonacci lattices. Ref: https://arxiv.org/abs/0912.4540 Input: N: [int]

Get indices of items matching the candidates patterns. Args: candidates_list: Single pattern or list of patterns (supports regex if use_r

Resolve a named teleoperation transform profile.

Resolve a reward function by name from a list or module path.

Get currently selected joint name

Get currently selected joint name

Return the currently selected joint name, if any.

ManiSkill TwoRobotPickCube reset: cube X=U(-0.05,0.05) Y=-0.15-U(0,0.1)+0.05; goal in air.

Reorder gains/angles according to the provided joint name list.

Joins all dimensions between dimensions (@begin_axis, @end_axis) into a flat dimension, for all tensors in nested dictionary or list or tuple

Load the entire objects_init_list.json and return as a dict.

Load the entire robots_init_list.json and return as a dict.

Copy a BC MLP's Linear weights into an rsl-rl ``ActorCritic`` actor. Args: actor_critic: an rsl-rl ``ActorCritic`` (or any module exposin

Build a launched MetaSim ``MujocoHandler`` around a LIBERO model_file. The LIBERO combined MJCF is fed to MetaSim's handler as the *scene* MJCF

ManiSkill PandaWristCam (panda_v3.urdf: 7 arm + 2 fingers + wrist camera link).

ManiSkill ``SimConfig`` (default tabletop) mapped onto the opt-in SimParamCfg recipe knobs.

Launch a MujocoHandler with the given scenario and step it. We bypass `_set_dof_targets` (which routes via per-robot joint names) and write d

Convert an MJCF XML file to a URDF XML file without using urdfpy.

Identical to mjlab.tasks.cartpole.cartpole_env_cfg.cartpole_smooth_reward().

My port's per-step reward BEFORE the dt scaling factor.

World quaternion (wxyz) of object ``name``.

Pad input tensor or array @seq in the time dimension (dimension 1). Args: seq (np.ndarray or torch.Tensor): sequence to be padded

Parse material/texture information from MuJoCo geom.

Parse material/texture information from MuJoCo geom.

Plot qpos[:n_show] overlaid for a (solid) vs b (dashed) + abs diff below.

Build homogeneous transforms (N,4,4) from batched pos (N,3) and optional quat (N,4).

Build homogeneous transforms (N,4,4) from batched pos (N,3) and optional quat (N,4). If quat is None, rotation is identity for all N.

Build homogeneous transforms (N,4,4) from batched pos (N,3) and optional quat (N,4).

Build homogeneous transforms (N,4,4) from batched pos (N,3) and optional quat (N,4).

Build homogeneous transforms (N,4,4) from batched pos (N,3) and optional quat (N,4).

VLA forward: returns (B,7) in metric/radian units (dpos, drot, gripper).

(H,W,3) uint8 -> (1,3,H,W) normalized float tensor.

Gravity vector in body frame. Shape ``(num_envs, 3)``.

Push an entry (dict of key->tensor). Clones tensor values on insert. Raises KeyError if any configured key is missing from the entry.

Robot qvel with the 2 finger DOFs dropped (panda).

Perform data randomization based on the given operation and distribution. Args: data: The data tensor to be randomized. Shape is (dim_0,

Rebase a LIBERO demo's embedded MJCF to local assets + make it dm_control-safe. Collector machine varies by suite (``/Users/yifengz`` for object/

Render an RGB frame for each qpos by forward-kinematics only (no stepping).

Replay qpos/qvel onto a fresh model and write an mp4. Returns out path or None.

Reset selected envs (defaults to all). ``seed`` is forwarded to the handler when supported, so the ``env.reset(seed=)`` contract reac