Pre-initialize trajectory tracking (before super().__init__()).

Pre-initialize trajectory tracking (before super().__init__()).

Apply action smoothing and apply pre-physics callbacks.

Apply pre-physics callbacks.

Make the active env importable as RoboTwin expects, in place. Two environment quirks have to be handled before importing any RoboTwin task, a

Map raw policy output to handler-ready action. Default: identity (clamp by handler-side joint limits, no scaling). Subclasses can imp

Quadratic sigmoid tolerance: 1 at x=0, 0 at ``|x|>=margin``.

Geodesic angle (rad) between two wxyz quaternions.

Randomize friction properties for IsaacGym simulator.

Randomize friction properties for IsaacGym simulator.

Randomize friction and restitution for MuJoCo simulator.

Randomize friction and restitution for MuJoCo simulator.

Randomize material properties for Newton simulator.

Rank demo-success episodes where BOTH native and shipped reach success. Scans the first ``max_scan`` demo-success episodes, keeps those where nat

Number of frames in the first episode of a v2 trajectory.

Bare reference mujoco: same model + init + ctrl as the MetaSim handler.

Register a NativeLiberoEnv subclass per vendored bundle: ``libero_native.<bundle>``.

Register a ``libero_plus_native.<dim>__<suite>__<stem>`` task per vendored bundle.

Rebase the bundle's embedded MJCF to local assets (overridable per family).

Replay one bridge trajectory in RoboVerse and measure achieved-qpos parity.

Replay one bridge trajectory; render video and/or measure object parity.

State trajectory from the raw reproduction recipe (recipe.build_replica).

Reset environments based on specified indices. Args: env_ids: List of environment ids which must be reset

Resolve geom IDs from a regex/name list. mjlab DR fns take ``asset_cfg.geom_names`` (a tuple of patterns); we accept either an ``asset_cfg``

Reward for bringing box to handover position.

Reward for grippers approaching the box.

Reward for bringing box to target position.

Reward for avoiding table collisions.

Pair each demo robot articulation key with the shipped task's robot name (order-aligned). Single-robot: ``[(demo_key, "panda")]``. Multi-robot (T

RoboTwin's per-task eval step budget (task_config/_eval_step_limit.yml). Matching this is part of evaluating "the same as RoboTwin": its own eval

RoboVerse object name from a RoboTwin actor name (no leading digit). The bridge disambiguates duplicate-named objects as ``name#1`` etc.; map the

Sample initial motion frame for ``n`` envs. Mjlab's adaptive sampler: bins are smoothed by a geometric kernel and weighted toward bin

Pick ``per_dim`` tasks for each (suite, perturbation-dimension) pair. Returns ``[(suite, task_id, task_name, category), ...]``. ``task_id`` is th

A smooth coordinated reach: both arms sweep toward the center cube. ``franka_right`` mirrors ``franka_left`` on joint 1 so the two arms move as a

Deterministic reach-down -> close gripper -> lift (7-D EE-delta + gripper).

Deterministic, non-trivial probe: press down + close gripper, then wiggle. Produces changing reaching/grasping reward and (sometimes) success, so

Set masses for IsaacGym simulator.

Set masses for IsaacGym simulator.

Apply domain randomization of start-up mode.

Setup IK solver for end-effector control.

Setup light randomizers for all lights.