MuJoCo Warp (MJWarp)

MJWarp is a GPU-optimized version of the MuJoCo physics simulator, designed for NVIDIA hardware.

Warning

MJWarp is in its Alpha stage, with many features still missing and limited testing so far.

MJWarp uses NVIDIA Warp to circumvent many of the sharp bits in MuJoCo MJX. Once MJWarp exits Alpha, it will be integrated into both MJX and Newton.

MJWarp is maintained by Google Deepmind and NVIDIA.

Installing for development

git clone https://github.com/google-deepmind/mujoco_warp.git
cd mujoco_warp
python3 -m venv env
source env/bin/activate
pip install --upgrade pip

During early development, MJWarp is on the bleeding edge - you should install Warp nightly:

pip install warp-lang --pre --upgrade -f https://pypi.nvidia.com/warp-lang/

Then install MJWarp in editable mode for local development:

pip install -e .

Now make sure everything is working:

pytest

Should print out something like XX passed in XX.XXs at the end!

Compatibility

The following features are implemented:

Category	Feature
Dynamics	Forward only
Transmission	JOINT, JOINTINPARENT
Actuator Dynamics	NONE
Actuator Gain	FIXED, AFFINE
Actuator Bias	NONE, AFFINE
Geom	PLANE, SPHERE, CAPSULE
Constraint	LIMIT_JOINT, CONTACT_PYRAMIDAL
Equality	Not yet implemented
Integrator	EULER, IMPLICITFAST
Cone	PYRAMIDAL
Condim	3
Solver	CG, NEWTON
Fluid Model	None
Tendons	Not yet implemented.
Sensors	Not yet implemented.

Benchmarking

Benchmark as follows:

mjwarp-testspeed --function=step --mjcf=humanoid/humanoid.xml --batch_size=8192

To get a full trace of the physics steps (e.g. timings of the subcomponents) run the following:

mjwarp-testspeed --function=step --mjcf=humanoid/humanoid.xml --batch_size=8192 --event_trace=True