Append new observation to buffer. Args: data: Observation tensor of shape (batch_size, ...).

Get device for testing, preferring CUDA if available. Can be overridden with FORCE_CPU=1 environment variable to test CPU-only behavior on GPU ma

Execute the sense pipeline: prepare -> graph -> finalize. Runs BVH refit, camera rendering, and raycasting in a single CUDA graph launch. Sho

Set joint position targets. Args: position: Target joint poisitions with shape (N, num_joints). joint_ids: Optional joint indices to

Initialize entity with simulation. Args: entity: Entity to initialize. device: Device to use ("cpu" or "cuda"). num_envs: Number of env

Set joint effort targets. Args: effort: Target joint efforts with shape (N, num_joints). joint_ids: Optional joint indices to set. If

Resolve names and IDs for all configured fields. This method ensures consistency between names and IDs for each field type. It handles three

Get the default value for a model field, caching for reuse. Returns the original values from the C MuJoCo model (mj_model), obtained from the

Create a scene and simulation from inline XML with sensors wired up. ``xml`` may be a single XML string (registered as the ``robot`` entity) or a

Set the joint state into the simulation. The joint state consists of joint positions and velocities. It does not include the root state.

Set joint velocity targets. Args: velocity: Target joint velocities with shape (N, num_joints). joint_ids: Optional joint indices to

Create entity with actuator from XML string. Args: xml_string: MuJoCo XML model string. actuator_cfg: Actuator configuration. Returns:

Load environment configuration for a task. Returns a deep copy to prevent mutation of the registered config.

Core randomization engine for model fields.

Create a color ramp from a base RGB color using parameter t. This function creates a color variation by adjusting saturation and value based on t

Set the root state into the simulation. The root state consists of position (3), orientation as a (w, x, y, z) quaternion (4), linear velocit

Call tick() with a controlled dt.

Append a new frame for all batch elements. Args: data: Tensor of shape (batch_size, ...).

Expand model fields to support per-environment parameters.

Load checkpoint. Extends the base implementation to: 1. Restore common_step_counter to preserve curricula state. 2. Migrate legacy checkp

Load XML content from fixture file. Args: fixture_name: Name of the fixture file (without .xml extension) or full path. Returns: XML con

Compute delayed observation for current step. Returns: Delayed observation with shape (batch_size, ...).

Helper to create a complete test environment with contact sensor. Sets up a scene with the specified entity and contact sensor configuration, com

Sample uniformly within a range. Args: lower: Lower bound of uniform range. upper: Upper bound of uniform range. size: Th

Multiply two quaternions together. Args: q1: The first quaternion in (w, x, y, z). Shape is (..., 4). q2: The second quaternion i

Compute low-level actuator control signal from high-level commands. Args: cmd: High-level actuator command. Returns: Control sig

Compile the underlying MjSpec into an MjModel.

Apply a quaternion rotation to a vector. Args: quat: The quaternion in (w, x, y, z). Shape is (..., 4). vec: The vector in (x, y,

Initialize an entity with a simulation.

Convert rotations given as quaternions to rotation matrices. Args: quaternions: The quaternion orientation in (w, x, y, z). Shape is (...

Convert rotations given as rotation matrices to quaternions. Args: matrix: The rotation matrices. Shape is (..., 3, 3). Returns:

Set the joint positions into the simulation. Like `write_joint_state_to_sim()` but only sets joint positions. Args: position: Tensor of

Return current frame (position, quaternion) from sim data.

Build a Warp Model with per-world mesh assignments. Args: spec: Scene spec (already merged with padded variant geoms). nworld: Number of si

Select GPUs based on CUDA_VISIBLE_DEVICES and user specification. This function treats the `gpu_ids` parameter as indices into the existing CUDA_

Advance actuator internal state by one physics substep. Called after each ``sim.step()`` within the decimation loop.

Print information message with color. Args: message: The message to print. color: Color name ('green', 'red', 'yellow', 'blue', 'cyan', 'ma

Apply an inverse quaternion rotation to a vector. Args: quat: The quaternion in (w, x, y, z). Shape is (..., 4). vec: The vector

Register an environment task. Args: task_id: Unique task identifier (e.g., "Mjlab-Velocity-Rough-Unitree-Go1"). env_cfg: Environment config

Compute env origins from sub-terrain origins. Args: num_envs: Number of environments to place. origins: Sub-terrain origins, shape ``

Extract command data for this actuator from entity data. Args: data: The entity data containing all state and target information. Retu

Prepare the entity for simulation after the spec has been compiled. Computes global index mappings, initializes actuators, and allocates all nwor

Convert rotations given as Euler angles in radians to Quaternions. Note: The euler angles are assumed in XYZ convention. Args:

Match a list of query regular expressions against a list of strings and return the matched indices and names. When a list of query regular expres

Write the scene XML and mesh assets to a directory. Creates ``scene.xml`` and an ``assets/`` subdirectory containing mesh files referenced by

Root + child body (hinge joint). Optional grandchild for arity tests.

Build entity from cfg kwargs and initialize it through the sim.

Darken an RGBA color by a given factor. Args: rgba: RGBA color tuple with values in range [0, 1]. factor: Darkening factor in range [0, 1],

Create rectangular border using four box geometries. Returns top, bottom, left, right boxes forming a hollow rectangle.

Save checkpoint. Extends the base implementation to persist the environment's common_step_counter and to respect the ``upload_model`` config

Set the root pose into the simulation. Like `write_root_state_to_sim()` but only sets position and orientation. Args: root_pose: Tensor

Set the root link (body origin) velocity into the simulation. Like `write_root_state_to_sim()` but only sets linear and angular velocity. Arg

Build a scene spec + variant_info from two variant spec_fns.

Clear all debug visualization queues.

Convert env_ids to consistent indexing format.

Assign worlds proportionally by weight (largest-remainder method). Returns a list of length *nworld* containing variant indices. Weights must be

Apply configuration settings to a compiled MjModel.