Search Ubiquitous advanced with the acquisition of MuJoCo
When you walk, your feet touch the ground. When you write, your fingers touch the pen. Physical connections are what make interaction with the world possible. However, for such a common occurrence, communication is a surprisingly complex phenomenon. Contacts can be soft or hard, bouncy or spongy, slippery or sticky, in microscopic scales at a two-body interface. No wonder our fingertips have four different types of touch sensors. This subtle complexity makes simulating physical contact—a vital component of robotics research—a challenging task.
A rich and powerful communication model to simulate physics has made MuJoCo a leading choice by robotics researchers and today, we’re proud to announce that as part of DeepMind’s mission to advance science, we’ve acquired MuJoCo and are making it freely available to everyone, to support research everywhere. Already widely used in the robotics community, including the chosen physical simulation for the DeepMind robotics team, MuJoCo features a rich communication model, a powerful scene description language, and a well-designed API. Together with the community, we will continue to improve MuJoCo as open source software under a permitted license. While we are working on preparing the code base, we are making MuJoCo freely available as a precompiled library.
A balanced model of communication. MuJoCo, which means molti-Atmosphereint with partnerntact, it hits a sweet spot with its contact form, which accurately and efficiently captures the salient features of object contact. Like other solid-body simulators, it avoids the fine details of deformations at the contact site, and often runs much faster than real time. Unlike other emulators, MuJoCo solves connection forces using the convex Gauss principle. Convexity guarantees unique solutions and well-defined inverse dynamics. The model is also flexible, providing multiple parameters that can be tuned to approximate a wide range of communication phenomena.
Real physics, no shortcuts. Because many simulators are initially designed for purposes such as gaming and cinema, they sometimes take shortcuts that prioritize stability over fidelity. For example, they may ignore gyroscopic forces or directly modulate velocities. This can be particularly harmful in the context of optimization: as artist and researcher Carl Sims first noted, the optimization agent can quickly detect and exploit these deviations from reality. In contrast, MuJoCo is a second order continuous time simulator, implementing the full motion equations. Familiar but non-trivial physical phenomena such as Newton’s cradle, as well as counterintuitive phenomena such as the Dzhanbekov effect, appear naturally. Ultimately, MuJoCo adheres closely to the equations that govern our world.
Portable code, clean API. MuJoCo’s core engine is written in pure C, which makes it easy to port to various architectures. The library produces deterministic results, with the scene description and simulated state completely encapsulated within two data structures. These make up all the information needed to recreate a simulation, including results from the intermediate stages, providing easy access to the internals. The library also provides fast and convenient calculations of commonly used quantities, such as Jacobean kinematic matrices and inertia matrices.
Description of a powerful scene. The MJCF scene description format uses cascading defaults—avoiding multiple repeated values—and contains elements for real-world robotic components such as equality constraints, motion capture tags, chords, actuators, and sensors. Our long-term roadmap includes standardizing MJCF as an open format to extend its usefulness beyond the MuJoCo ecosystem.
Biomechanical simulation. MuJoCo includes two powerful features that support both human and animal musculoskeletal models. The spatial orientation of tendons, including wrapping around bones, means that applied forces can be correctly distributed to joints, describing complex effects such as the variable torque lever in the knee enabled by the tibia. The MuJoCo muscle model captures the complexity of biological muscles, including activation states and force-length-velocity curves.
newly PNAS A perspective that explores the state of simulation in robotics identifies open source tools as essential to advancing research. The authors’ recommendations are to develop and validate open-source simulation platforms as well as to create open, community-curated libraries of validated models. In line with these goals, we are committed to developing and maintaining MuJoCo as a free, community-driven open source project with best-in-class capabilities. We’re currently working hard on getting MuJoCo fully open sourced, and encourage you to download the software from the new homepage and visit the GitHub repository if you’d like to contribute. Email us if you have any questions or suggestions, and if you’re also passionate about pushing the boundaries of realistic physics simulation, we’re hiring. We can’t promise that we’ll be able to tackle everything right away, but we’re eager to work together to make MuJoCo the physics simulator we’ve all been waiting for.
MuJoCo at DeepMind. Our robotics team uses MuJoCo as a simulation platform for many projects, often via our dm_control Python stack. In the carousel below, we highlight a few examples to showcase what can be emulated in MuJoCo. Of course, these clips represent only a small part of the vast possibilities for how researchers can use the simulator. For higher quality versions of these clips, please click here.