Reset the motion command (similar to `CommandTerm.reset()` in Isaac Lab).

Reset the specified environments. ``seed`` is forwarded to the handler when supported (this base reimplements ``reset`` and does not

Match a list of regular expressions in a dictionary against a list of strings and return the matched indices, names, and values. If the :attr:`pr

Per-joint offset tensor. Mjlab semantic: when ``use_default_offset=True``, offset comes from the entity's default-joint-position map (``defau

Per-joint scale tensor for vectorized action processing.

Retarget 21x3 hand landmarks to a 20-joint Wuji target.

Roll out one episode of a ManiSkill env and capture full state. Returns a `ManiskillRollout` ready for diffing against a MetaSim run.

Convert rotation matrix to quaternion (w, x, y, z).

Generate samples using the flow network. Args: model: The flow network. shape: Shape of the output tensor (b

Save current episode to the episodes list

Set the action to execute. The action can be low-level control signals or high-level abstract commands.

One-period sinusoid; deterministic across runs.

ManiSkill's kinematic table box; its top surface sits at z = 0.

Recursively convert tensors to lists/numpy arrays.

Convert a string to CamelCase. E.g., 'Sauce Pyramid' -> 'SaucePyramid'.

Convert a string to snake_case. E.g., 'Sauce Pyramid' -> 'sauce_pyramid'.

Apply transform to the mesh based on the origin element in URDF.

Transform points to new coordinates with transformation matrix. Input: cloud: [torch.FloatTensor, (N,3)] points

Process pending events and return whether the window remains open.

Advance / retreat ``terrain_levels`` per-env (mjlab parity).

Validate a two-hand teleoperation packet.

arm_pd_ee_target_delta_pose_align2 + gripper_pd_joint_pos (the deployed Bridge mode).

Call the query.

Initializes the model. Parameters: backbones: torch module of the backbone to be used. See backbone.py transformer: t

A fixed, deterministic action sequence: 5 zero warmup steps then a small sinusoid (mirrors the LIBERO eval warmup, kept identical across both envs

Append a flat ground geom + a light.

Return (qpos, ctrl) that match mjlab's standing keyframe.

For each (joint_name, kp, kv, effort_limit, armature), wire a PD actuator. mjlab's `BuiltinPositionActuatorCfg` maps to: * MuJoCo `general`

Adjust tensors from checkpoint_state to better match model.state_dict() shapes. - If a tensor differs only in the leading (0-th) dimension and th

Build one agent's name-keyed v2 demo (init_state + per-step states/actions).

Build a pair of RolloutResults restricted to matching joints (by suffix). The MetaSim path prefixes joints with `<robotname>/`, so we strip prefi

Resolve the ALOHA-AgileX URDF via the RoboTwin locator (clone -> mirror -> HF).

Push processed action into the simulator handler. Default: ``handler.set_dof_targets`` (joint-position control, the most common case

Set ManiSkill's high-friction grasp material on every robot's fingers (once, idempotent). ManiSkill applies this material in code rather than

Apply ManiSkill's gripper grasp material to every robot's fingers (once, idempotent).

Apply joint2 lift control when object is grasped.

Apply joint2 lift control when object is grasped.

Apply joint2 lift control when object is grasped.

Copy compiled-model placement fields robosuite mutated at reset.

Set the per-object contact materials a ManiSkill task customizes in code (once, idempotent). Mirrors task-build code that gives an object a n

Simple gripper control: close when near object.

Simple gripper control: close when near object.