MuJoCo Playground
Google DeepMind / Playground contributors
GPU-accelerated robot-learning environments, including tasks for dexterous manipulation.
What you can explore
Explore training and evaluation workflows using the documented environments and robot models.
Before you use it
Check GPU, software and model requirements for the chosen environment. Simulation success does not establish performance on an arbitrary real hand.
Original contributors
Kevin Zakka and collaborators
License / access: Code: Apache 2.0 · robot assets retain their own licenses
Start with the right setup.
- Environment
- MuJoCo / MJX
- Embodiment
- Task-specific robot assets
What you need
Follow the JAX, MJX and accelerator setup for the selected environment.
First useful check
Inspect a released manipulation environment before starting training.
Where it stops
Code is Apache 2.0; assets have separate terms. A released task is not universal hardware support.
Editorial reference · checked 2026-09-22