Open-sourcing MuJoCo

by

In October 2021, we introduced that we acquired the MuJoCo physics simulator, and made it freely out there for everybody to assist analysis all over the place. We additionally dedicated to growing and sustaining MuJoCo as a free, open-source, community-driven venture with best-in-class capabilities. In the present day, we’re thrilled to report that open sourcing is full and the complete codebase is on GitHub! 

Right here, we clarify why MuJoCo is a superb platform for open-source collaboration and share a preview of our roadmap going ahead.

A platform for collaboration

Physics simulators are essential instruments in fashionable robotics analysis and sometimes fall into these two classes: 

  1. Closed-source, industrial software program.
  2. Open-source software program, typically created in academia. 

The primary class is opaque to the person, and though generally free to make use of, can’t be modified and is tough to grasp. The second class typically has a smaller person base and suffers when its builders and maintainers graduate.

MuJoCo is likely one of the few full-featured simulators backed by a longtime firm, which is really open supply. As a research-driven organisation, we view MuJoCo as a platform for collaboration, the place roboticists and engineers can be part of us to develop one of many world’s greatest robotic simulators.

Options that make MuJoCo significantly engaging for collaboration are:

  • Full-featured simulator that may mannequin advanced mechanisms. 
  • Readable, performant, moveable code.
  • Simply extensible codebase.
  • Detailed documentation: each user-facing and code feedback.

We hope that colleagues throughout academia and the OSS neighborhood profit from this platform and contribute to the codebase, enhancing analysis for everybody.

Efficiency

As a C library with no dynamic reminiscence allocation, MuJoCo could be very quick. Sadly, uncooked physics velocity has traditionally been hindered by Python wrappers, which made batched, multi-threaded operations non-performant because of the presence of the International Interpreter Lock (GIL) and non-compiled code. In our roadmap beneath, we handle this subject going ahead. 

For now, we’d wish to share some benchmarking outcomes for 2 widespread fashions. The outcomes have been obtained on a regular AMD Ryzen 9 5950X machine, operating Home windows 10.

These values have been obtained from our testspeed pattern code. Notably, management noise is injected into the actuators stopping the system from settling into a set state, and are due to this fact consultant of real-world efficiency.

Roadmap

Right here’s our near-term roadmap for MuJoCo:

  • Unlock MuJoCo’s velocity potential with batched, multi-threaded simulation.
  • Assist bigger scenes with enhancements to inner reminiscence administration.
  • New incremental compiler with higher mannequin composability.
  • Assist for higher rendering by way of Unity integration.
  • Native assist for physics derivatives, each analytical and finite-differenced.

Study extra

Useful assets about MuJoCo:

We look ahead to receiving your contributions!

Leave a Comment