Advancing Research Everywhere with the Acquisition of MuJoCo
Physical contact is a complex phenomenon that plays a crucial role in our interactions with the world. To accurately simulate physical contact in robotics research, a robust contact model is needed. That’s why the MuJoCo physics simulator has become a leading choice for robotics researchers. We’re excited to announce that DeepMind has acquired MuJoCo and is making it freely available to support research everywhere.
MuJoCo boasts a rich contact model, powerful scene description language, and a well-designed API. It is widely used within the robotics community, including by DeepMind’s own robotics team. As part of our mission to advance science, we will continue to improve MuJoCo as open-source software under a permissive license. While we work on preparing the codebase, we are making MuJoCo available as a precompiled library.
The MuJoCo contact model accurately and efficiently captures the salient features of contacting objects. It avoids the fine details of deformations at the contact site, running much faster than real-time. The model resolves contact forces using the convex Gauss Principle, ensuring unique solutions and well-defined inverse dynamics. MuJoCo’s flexibility allows researchers to tune parameters to approximate a wide range of contact phenomena.
Unlike other simulators, MuJoCo implements the full Equations of Motion, without taking shortcuts that prioritize stability over accuracy. This second-order continuous-time simulator faithfully adheres to the physics governing our world. It can naturally reproduce familiar physical phenomena like Newton’s Cradle and even unintuitive ones like the Dzhanibekov effect.
The core engine of MuJoCo is written in pure C, making it easily portable to various architectures. The library produces deterministic results and provides fast computations of commonly used quantities. The MJCF scene-description format includes elements for real-world robotic components, and our roadmap includes standardizing it as an open format.
MuJoCo also supports musculoskeletal models of humans and animals. It accurately captures effects like tendon routing and the complexity of biological muscles, enabling researchers to study biomechanics.
Open-source tools are crucial for advancing research in robotics, and MuJoCo will be developed and maintained as a free, open-source, community-driven project. We’re currently preparing MuJoCo for full open sourcing and encourage researchers to download the software and contribute to its development on GitHub.
DeepMind’s robotics team has already used MuJoCo for various projects and showcases its capabilities in simulation. These examples represent just a fraction of what can be achieved using MuJoCo. By making MuJoCo freely available, we hope to push the boundaries of realistic physics simulation and accelerate advancements in robotics research.
Are you interested in joining us to work on exciting projects like MuJoCo? We’re hiring and eager to collaborate with passionate individuals who want to contribute to the future of robotics.
This is an exciting step towards advancing research with the acquisition of MuJoCo. Stay tuned for more updates as we continue to develop and enhance this powerful physics simulator.