mujoco-warp

library3.9.0.1pypypiunverified

MuJoCo Warp (MJWarp) is a differentiable simulation framework built on MuJoCo and Google's Warp library for GPU-accelerated physics. It enables gradient-based optimization, reinforcement learning, and trajectory optimization with full differentiability. Current version 3.9.0.1, requires Python >=3.10. Active development.

ai-ml

pip install mujoco-warp

github ↗homepage ↗

INSTALL

IMPORT

SIG · MUJOCO-WARP

mujoco-warp

ai-mlpythonv3.9.0.1

harness data pending

Install & Compatibility

Where this runs

No compatibility data collected yet for this library.

Code

Verified usage

Verified import paths — ran on the pinned version, not inferred.

Simulator

✓ from mujoco_warp import Simulator

✗ from mujoco_warp import Simulator

Model

✓ from mujoco_warp import Model

Data

✓ from mujoco_warp import Data

Initialize a Simulator from an MJCF string, step through physics, and print joint positions.

import mujoco_warp from mujoco_warp import Simulator sim = Simulator.from_xml(''' <mujoco> <worldbody> <freejoint/> <geom type="sphere" size="0.1" pos="0 0 0.5"/> </worldbody> </mujoco> ''') sim.reset() for _ in range(100): sim.step() print(sim.data.qpos)

Debug

Known issues

breakingAPI overhaul in v3.0: The old `WarpEnv` class was removed; use `Simulator` instead. State access changed from `sim.state()` to `sim.data`.

fix

Upgrade code: use `from mujoco_warp import Simulator`, and access state via `sim.data.qpos`, `sim.data.qvel`.

affects: <3.0

deprecated`mjcf.load` is deprecated in v3.9.0. Use `Simulator.from_xml` or `Simulator.from_mjcf` instead.

fix

Replace `mjcf.load(xml_string)` with `Simulator.from_xml(xml_string)`.

affects: >=3.9.0, <3.9.1

gotchaWarp kernels must be compiled on first call, causing long startup time. This is not a bug but a performance characteristic.

fix

Warm up by calling `sim.step()` once before timing critical loops.

affects: all

gotchaThe `Simulator` does not support multiple environments in one instance. For batched simulation, use `mujoco_warp.batch.SimulatorBatch` (experimental).

fix

For batched sim, import `from mujoco_warp.batch import SimulatorBatch` and pass `num_envs`.

affects: all

Upgrade

Version history

3.9.0.1latest on PyPI · released May 27, 2026

Audit

Dependencies

mujocorequiredCore physics engine dependency

warp-langrequiredGPU kernel library for differentiable simulation

dm-robotics-geometryoptionalGeometry utilities (e.g., transformations)

Agent activity

17 hits · last 30 days

node

Resources

githubhttps://github.com/google-deepmind/mujoco_warp ↗

homepagehttps://github.com/google-deepmind/mujoco_warp ↗