In October 2021, the acquisition of the MuJoCo physics simulator was announced. The simulator is now available for free on GitHub, marking the completion of the open sourcing process.
MuJoCo is a valuable platform for open-source collaboration in robotics research. Unlike closed-source commercial software or smaller, less sustainable academic projects, MuJoCo is fully open-source and backed by an established company. The goal is to develop one of the best robot simulators in the world through collaboration with roboticists and engineers.
Some key features that make MuJoCo attractive for collaboration include its ability to model complex mechanisms, its readable and performant code, and its extensibility. Additionally, detailed documentation is available to assist users.
In terms of performance, MuJoCo is very fast due to its C library design. However, the presence of the Global Interpreter Lock and non-compiled code in Python wrappers has historically limited batched, multi-threaded operations. The roadmap addresses this issue and plans to improve the simulator’s speed potential going forward. Benchmarking results on common models have been provided to showcase its performance.
Looking ahead, the near-term roadmap for MuJoCo includes unlocking its speed potential through batched, multi-threaded simulation, improving internal memory management for larger scenes, introducing a new incremental compiler for better model composability, integrating with Unity for enhanced rendering, and supporting physics derivatives.
For those interested in learning more about MuJoCo, helpful resources are available. Contributions from the academia and open-source community are welcomed and encouraged.
Overall, the open sourcing of MuJoCo provides a powerful platform for collaboration in robotics research and aims to enhance research capabilities for everyone involved.