myosuite-mjx

GPU-Accelerated Musculoskeletal Simulation

MyoSuite is the standard platform for research on muscle-driven, biomechanically realistic control (e.g. the MyoHand), simulated in MuJoCo and wrapped in the Gym API. Its original CPU-based backend caps how many environments can be stepped in parallel — the binding constraint for large-scale reinforcement learning.

myosuite-mjx gives MyoSuite an MJX (MuJoCo-XLA) backend so simulation runs on GPU/TPU inside JAX, stepping thousands of environments in parallel with the whole rollout kept on-device. This work is done through Micropilot, in collaboration with MyoSuite creator Vittorio Caggiano (FAIR, Meta).

What It Enables

• Porting MyoSuite’s musculoskeletal environments to MJX / JAX.
• Preserving muscle-tendon dynamics and contact-rich simulation fidelity.
• Compatibility with modern MuJoCo and the standard Gym API.
• GPU-accelerated, massively parallel rollouts for large-scale RL of dexterous control.

It is the simulation engine behind the rest of my robotics stack — the same MyoHand model that supplies tendon-control targets in emg2tendon becomes trainable at scale once it runs on MJX.

Code: github.com/micropilot/myosuite-mjx · Read the blog post. </content>