Mock env/sensors, verify subsampling + concatenation.

Custom preset is registered in ALL_TERRAIN_PRESETS.

TextureCfg should add textures to spec.

Test texture extraction with a simple textured model.

I^2R heating raises T, which raises effective resistance and decays torque over time.

Points on z = 0.5*x plane → normal perpendicular to that.

TorchArray supports PyTorch operations.

TorchArray modifications affect underlying warp array.

No-state-estimation tasks remove observations that depend on state estimation.

Tracking play tasks should disable RSI randomization.

Tracking play tasks should use sampling_mode='start'.

use_artifact is NOT called when using the tensorboard logger.

use_artifact is called for both legacy 'wandb' and current 'WandbLogWriter' logger types. Regression test: rsl-rl-lib 5.4 renamed the WandB logger

All tracking tasks should have a 'motion' command of type MotionCommandCfg.

All tracking tasks should have a self_collision sensor.

Training mode tasks should have observation corruption enabled for policy.

Tuple names with matching IDs should validate without error.

Test that two identical actuated entities write controls to the correct global positions.

Each target gets one <position> + one <velocity>, in halves.

Ensure all tyro.cli() calls include config=mjlab.TYRO_FLAGS.

Regression test for #677 due to incompatibilities between declared types and default values of the env configs.

Test conversion with Unitree G1 robot.

Test conversion with Unitree Go1 robot.

Test that an unnamed freejoint is auto-named during entity init.

Test that update_history=False doesn't modify the buffer.

Verify that update_period controls lag update frequency.

Lags update only at scheduled intervals.

Empty worldbody fails with a clear root-body message.

Error messages use the standardized mjlab.entity prefix and Hint suffix.

Delayed values correctly match current lags.

Reordering the variant dict does not change per-variant per-world fields.

Velocity actuator compiles for continuous joints with no range.

Velocity actuator force is bounded by effort_limit.

P-only velocity tracking: tau = K * kp * (target - qdot) / R.

Zero pos target, joint at rest: qfrc = kd * vel_target.

All velocity tasks should have a velocity command.

viscous_damping=None preserves XML joint damping (0.8).

Explicit viscous_damping overrides XML damping.

Explicit 0.0 overrides XML damping (distinguishes None from 0.0).

Nonzero viscous_damping produces a dissipative passive force.

Viser dynamic meshes must not collapse all worlds onto env0's mesh.

Convex-hull handles must differ across variants, not all show env0's hull.

Per-world body_mass matches independent compile when variants have visual+collision split. This verifies the slot-driven reference compile preserve

At rest, ctrl = V -> tau = K * V / R.

Docstring promises ``voltage_limit=0`` disables clamping. Verify against MuJoCo's ``dcmotor_voltage`` (which only clamps when ``Vmax > 0``).

Wildcard matching joints should warn about unactuated tendons.

Perfectly upright robot on flat ground → reward ≈ 1.

30° roll → reward significantly below 1.

Asset keys with path traversal must not escape the output directory.

COM velocity write must produce the same qvel as manual conversion.

write() only includes assets referenced in the generated XML.

Test write_root_velocity uses qpos (not stale xquat). Writing pose then velocity without forward() must work. This would fail if write_root_veloc

Test exporting scene to zip file.

Auto-detection raises for non-standard actuators when command_field is None.

Explicit command_field allows non-standard XML actuators.

XmlActuator raises error when no targets have matching XML actuators.

XmlActuator filters to joints with XML actuators when using wildcard.

XML actuator matches targets through internal attach prefix.

Builtin actuator preserves XML armature/frictionloss when not set.

Verify sensors defined in entity XML are automatically discovered and exposed.

sim.step() leaves xpos stale; sim.forward() makes it match qpos. In MuJoCo, mj_step = mj_step1 (forward kinematics + forces) + mj_step2 (integrat

Test that delay with min_lag=max_lag=0 returns current observation.

Terminate when the episode length exceeds its maximum.

Converts a string from snake case to camel case. Args: snake_str: A string in snake case (i.e. with '_') to: Convention to conver

Converts a string from camel case to snake case. Args: camel_str: A string in camel case. Returns: A string in snake case (i

Convert to tuple representation.

Deprecated. Use ``write(output_dir, zip=True)`` instead.

Write the MjSpec to a zip file.

Reward heading error for heading-controlled envs, angular velocity for others. The commanded xy angular velocities are assumed to be zero.

Reward for tracking the commanded base linear velocity. The commanded z velocity is assumed to be zero.

Get all tracking task IDs.

r"""Transform input points in a given frame to a target frame. This function transform points from a source frame to a target frame. The transfor

Transforms poses from one coordinate frame to another preserving relative poses. Args: src_poses: Input pose sequence (shape [T, 4, 4]) f

Transmission type of this actuator.

r"""Un-project depth image into a pointcloud. This function converts orthogonal or perspective depth images into points given the calibration mat

De-normalizes a given input tensor from range of [-1, 1] to (lower, upper). .. note:: It uses pytorch broadcasting functionality to deal

Get the unwrapped environment.

Get the unwrapped environment (base case for wrapper chains).

Update renderer with simulation data.

Push new data and refresh visible plots.

Push latest camera images/frustums to GUI.

Push new values and re-render the bars. Args: terms: List of ``(term_name, value_array)`` tuples.

Update actuator state after a simulation step. Base implementation does nothing. Override in subclasses that need per-step updates. Args