Configuration for a scene entity that is used by the manager's term. This configuration allows flexible specification of entity components either b

Configuration for an observation term. Processing pipeline: compute → noise → clip → scale → delay → history. Delay models sensor latency. Histor

Configuration for an observation group. An observation group bundles multiple observation terms together. Groups are typically used to separate o

Reference to a MuJoCo object (body, joint, site, etc.). Used to specify which object a sensor is attached to and its frame of reference. The enti

GPU-accelerated MuJoCo simulation powered by MJWarp. CUDA Graph Capture ------------------ On CUDA devices with memory pools enabled, the simul

An entity represents a physical object in the simulation. Entity Type Matrix ================== MuJoCo entities can be categorized along two di

Manages observation computation for the environment. The observation manager computes observations from multiple terms organized into groups. Eac

Configuration for a simulation scene.

Configuration for a reward term.

Specifies what to match on one side of a contact. Args: mode: MuJoCo element type to match against ("geom", "body", or "subtree"). pattern:

Configuration for MuJoCo built-in position actuator. Under the hood, this creates a <position> actuator for each target and sets the stiffness, d

Entity config for per-world mesh variants. Provide a dict of named variants (each value is a callable returning an ``MjSpec``) and optionally an

Configuration for an event term. Event terms trigger operations at specific simulation events. They're commonly used for domain randomization, st

Configuration for a :class:`ContactSensor`. A contact sensor watches contacts between a ``primary`` set of elements and an optional ``secondary``

Raycast sensor configuration. Supports multiple ray patterns (grid, pinhole camera, ring) and alignment modes.

Manager-based RL environment.

Grid pattern - parallel rays in a 2D grid.

Configuration for a recorder term. ``func`` must be a :class:`RecorderTerm` subclass. Function-based terms are not supported because recorder ter

Configuration for terrain as an entity.

Serve stochastically delayed observations from a rolling history. Wraps a CircularBuffer to simulate observation delays by returning frames from T-

Configuration for a metrics term. Attributes: per_substep: If True, evaluate this term once per physics substep inside the decimation loop

Configuration for MuJoCo simulation parameters.

Configuration for MuJoCo built-in motor actuator. Under the hood, this creates a <motor> actuator for each target and sets its effort limit and g

Configuration for a manager-based RL environment. This config defines all aspects of an RL environment: the physical scene, observations, actions

Configuration for a camera sensor. A camera sensor can either wrap an existing MuJoCo camera (``camera_name``) or create a new one at the specifi

Minimal concrete viewer for testing.

Orchestrates recorder terms during environment rollouts. Holds a collection of :class:`RecorderTerm` instances and calls their lifecycle methods

Fixed-length circular buffer for batched tensor history. Stores history with shape (max_len, batch_size, ...) internally. The `buffer` property r

Accumulates per-step metric values, reports episode averages. Unlike rewards, metrics have no weight, no dt scaling, and no normalization by epis

Configuration for a termination term.

Config for a single neural network model (Actor or Critic).

Configuration to modify collision properties of geoms in the MuJoCo spec. Supports regex pattern matching for geom names and dict-based field resol

Native MuJoCo ``<dcmotor>`` wrapper. Models a DC motor: torque is derived from voltage via the motor constant K and back-EMF, tau = K * (V - K *

Wrap existing XML-defined actuators. Detects the actuator type (position, velocity, motor, muscle) from the MuJoCo spec element during ``edit_spe

Manages reward computation by aggregating weighted reward terms. Reward Scaling Behavior: By default, rewards are scaled by the environment ste

Wraps mjwarp objects to expose Warp arrays as PyTorch tensors. Automatically converts Warp array attributes to TorchArray objects on access, enab

Configuration for a curriculum term. Curriculum terms modify environment parameters during training to implement curriculum learning strategies (

Configuration for differential IK action space. Converts task-space commands (position and/or orientation) into joint-space targets via damped le

Configuration for a resolvable entity field.

Configuration for MLP-based learned actuator model. This actuator learns the mapping from joint commands and state history to actual torque outpu

Configuration for NaN guard.

Configuration for DC motor actuator with velocity-based saturation. This actuator implements a DC motor torque-speed curve for more realistic act

Electric actuator parameters.

Configuration for ideal PD actuator.

Configuration for sampling flat patches on a heightfield surface.

Config for the PPO algorithm.

Terrain entity. The terrain is a grid of sub-terrain patches (num_rows x num_cols), each with a spawn origin. When num_envs exceeds the number of

Manages a single batched mesh handle with lazy mesh creation.

Continuous space with optional bounds.

A domain randomization sampling distribution. Built-in distributions (``uniform``, ``log_uniform``, ``gaussian``) are provided as module-level in

Manages event-based operations for the environment. The event manager triggers operations at different simulation events: startup (once at initia

Configuration for joint position control.

Base runner that persists environment state across checkpoints.

A domain randomization operation. Built-in operations (``abs``, ``scale``, ``add``) are provided as module-level instances. Users can create cust

Configuration for joint position control relative to current positions. target = current_joint_pos + action * scale The ``offset`` field inherit

Structural identity of a mesh slot in the merged template.

Generates procedural terrain grids with configurable difficulty. Creates a grid of terrain patches where each patch can be a different terrain ty

Warp array that behaves like a torch.Tensor with shared memory.

Wraps an environment to record video during interaction. A minimal wrapper that records frames as the environment steps. Delegates all attribute

Interactive Viser-based viewer with playback controls.

Minimal env-like object for testing observation functions.

ONNX-exportable model that wraps the policy and bundles motion reference data.

Manages action processing for the environment. The action manager aggregates multiple action terms, each controlling a different entity or aspect

High-level actuator command with targets and current state. Passed to actuator's `compute()` method to generate low-level control signals. All te

Implicit-integration version of IdealPdActuator. Both consume a position target and a velocity target with kp/kd gains. The difference is in how

HSV color representation.

ViserMujocoScene with debug visualization and warp tensor conversion. Adds debug primitives (arrows, ghosts, spheres, cylinders, ellipsoids, coor

No-op fallback used when no recorder terms are configured. All methods are no-ops. This class is not a :class:`ManagerBase` subclass so it carrie

Raycast sensor output data. Note: Fields are views into GPU buffers and are valid until the next ``sense()`` call.

Manages termination conditions for the environment. The termination manager aggregates multiple termination terms to compute episode done signals

Minimal mock of a motion object with tensor attributes.

Update a reward term's weight and/or params based on training steps. Each stage specifies a ``step`` threshold and optionally a ``weight`` and/or

Camera sensor output data. Shapes: - rgb: [num_envs, height, width, 3] (uint8) - depth: [num_envs, height, width, 1] (float32) - segmen

Holds checkpoint discovery and loading callbacks for the viewer.

Manages curriculum learning for the environment. The curriculum manager updates environment parameters during training based on agent performance

Configuration to add a material to the MuJoCo spec. Optionally assigns the material to geoms matching ``geom_names_expr``.

Debug visualizer for MuJoCo's native viewer. This implementation directly adds geometry to the MuJoCo scene using mjv_addGeoms and other MuJoCo v

Placeholder for absent metrics manager that safely no-ops all operations.

Configuration for tendon length control.

RayCastSensor that reports per-frame vertical clearance above terrain. Inherits all raycasting configuration from :class:`RayCastSensorCfg`. The se