Yuval Tassa
86e9860106
Replaces the box-box collider's manifold generation and post-filtering with a single structured implementation, and deletes the accumulated repair logic it obsoletes. Net 319 lines out of the engine. Algorithm: - The separating-axis test keeps the closed-form support evaluation and chooses the axis of maximum separation among the 15 candidates by plain argmax. Edge-cross axes whose cross product has norm below rounding are skipped: in the nearly-parallel regime their direction is cancellation noise, previously the source of arbitrary-normal contacts with box-scale spurious depth. A winning edge axis within eight degrees of the best face axis is replaced by that face unless it is better by five percent (ODE's classic fudge): resting stacks otherwise flip between the edge and face contact codes by rounding noise from step to step, thrashing the solver warm start until the stack explodes. The substitution runs after the search rather than filtering during it, so a worse non-aliasing edge cannot steal the contact the substitution meant to give to the face. - Face contacts clip the incident face against the reference face's side planes (Sutherland-Hodgman). Depth is measured along the reference normal only, never as a Euclidean distance between unrelated points. Contact position is midway between the surfaces along the normal, so its distance to either box is bounded by half the contact depth. Every surviving vertex of the clipped polygon becomes a contact, so the manifold is the actual contact patch, at most eight points as before. - Edge contacts use the closest-point pair between the two supporting edge segments. A near-zero axis component makes the support-corner sign ambiguous; both signs are enumerated and the closest witness pair wins. - Margin is an acceptance band throughout: SAT early-out and clip acceptance. - The rounding thresholds are stated per precision. The separation tests are the ones that cost correctness: comparing exactly against the margin reports a pair overlapping by less than the rounding error of its own support evaluation as separated, and the boxes pass through each other. Over 239k overlapping pairs that is eight misses under mjUSESINGLE and none in double; the collider this replaces misses the same eight. Slack proportional to the summed half-sizes leaves five, which overlap by 7e-9 to 3e-8 of their own scale, below single-precision epsilon, where the boxes are not distinguishable from touching. Erring toward contact is the safe direction: the driver already excludes a contact whose distance reaches the margin. Deleted: the conditional acceptance cascade keyed on how many points earlier generators emitted, the u/v clamping that fabricated contacts from out-of-range projections, the outside-box removal filter and its missing-fallback hole, exact-floating-point deduplication, and the edge-path depth clamp. The structure makes those bug classes unrepresentable rather than filtered: depth is a projection by construction. Every reported depth is the exact support overlap along the contact's own normal, verified over 246k overlapping poses to within two ulps; the face preference costs direction, not depth, deviating from the minimum-translation axis by at most 8.1 degrees and 5.3% of its depth. The previous implementation is preserved verbatim as mjc_BoxBoxLegacy in test/engine/boxbox_legacy.c, a static library that only the box-box tests link, so the claims above are measured rather than asserted. It needs no private engine symbols. Three tests compare against it: - NearAlignedManifoldIsExact sweeps the relative angle of a resting pair across the regime where the edge-cross axes degenerate into noise, pinning the full clipped polygon and a contact normal equal to the face normal exactly, where the previous collider drifts off it. - AlignedTowerStands settles a twenty-box tower, which comes to rest four million times quieter than under the previous collider, which never settles and eventually topples. - ShallowOverlapSurvivesRounding pins a pair overlapping by 7e-8 of its scale, reported as separated under mjUSESINGLE without slack on the separation tests. On stacks of plates across aspect ratios from 4:1 to 25:1, five layouts each, the collider settles into a tight band of 1e-4 to 3e-4 while the previous one intermittently blows up to as much as 2.6e-2. engine_collision_box_fuzz_test.cc cross-validates randomized poses against GJK/EPA on identical box meshes and against a spherical-Fibonacci support sweep, with hard gates per sample: no phantom penetration, no missed contact at zero margin, no contact deeper than the true depth, contacts within half their own depth of both boxes, and one normal per manifold. Both invocations run in about a second. EdgeContactAtDepthBound's tolerance widens to the five percent design band; the three-orders-of-magnitude depth bug it pins is still caught, the deviation being 0.13 percent of the depth. The 100-box pile benchmark steps about 7% faster with 1.6% fewer contacts. PiperOrigin-RevId: 965114952 Change-Id: Ie98cdcce8d1aed3ff2da938cb29703fd9c241258
MuJoCo stands for Multi-Joint dynamics with Contact. It is a general purpose physics engine that aims to facilitate research and development in robotics, biomechanics, graphics and animation, machine learning, and other areas which demand fast and accurate simulation of articulated structures interacting with their environment.
This repository is maintained by Google DeepMind.
MuJoCo has a C API and is intended for researchers and developers. The runtime simulation module is tuned to maximize performance and operates on low-level data structures that are preallocated by the built-in XML compiler. The library includes interactive visualization with a native GUI, rendered in OpenGL. MuJoCo further exposes a large number of utility functions for computing physics-related quantities.
We also provide Python bindings and a plug-in for the Unity game engine.
Documentation
MuJoCo's documentation can be found at mujoco.readthedocs.io. Upcoming features due for the next release can be found in the changelog in the "latest" branch.
Getting Started
There are two easy ways to get started with MuJoCo:
-
Run
simulateon your machine. This video shows a screen capture ofsimulate, MuJoCo's native interactive viewer. Follow the steps described in the Getting Started section of the documentation to getsimulaterunning on your machine. -
Explore our online IPython notebooks. If you are a Python user, you might want to start with our tutorial notebooks running on Google Colab:
- The introductory tutorial teaches MuJoCo basics:
- The Model Editing tutorial shows how to create and edit models procedurally:
- The rollout tutorial shows how to use the multithreaded
rolloutmodule: - The LQR tutorial synthesizes a linear-quadratic controller, balancing a
humanoid on one leg:
- The least-squares tutorial explains how to use the Python-based nonlinear
least-squares solver:
- The MJX tutorial provides usage examples of
MuJoCo XLA, a branch of MuJoCo written in JAX:
- The differentiable physics tutorial trains locomotion policies with
analytical gradients automatically derived from MuJoCo's physics step:
Installation
Prebuilt binaries
Versioned releases are available as precompiled binaries from the GitHub releases page, built for Linux (x86-64 and AArch64), Windows (x86-64 only), and macOS (universal). This is the recommended way to use the software.
Building from source
Users who wish to build MuJoCo from source should consult the build from
source section of the documentation. However, note that the commit at
the tip of the main branch may be unstable.
Python (>= 3.10)
The native Python bindings, which come pre-packaged with a copy of MuJoCo, can be installed from PyPI via:
pip install mujoco
Note that Pre-built Linux wheels target manylinux2014, see
here for compatible distributions. For more
information such as building the bindings from source, see the Python bindings
section of the documentation.
Versioning
We aim to release MuJoCo in the first week of each month. Our versioning standards changed to modified Semantic Versioning in 3.5.0, see versioning for details.
Contributing
We welcome community engagement: questions, requests for help, bug reports and feature requests. To read more about bug reports, feature requests and more ambitious contributions, please see our contributors guide and style guide.
Asking Questions
Questions and requests for help are welcome as a GitHub "Asking for Help" Discussion and should focus on a specific problem or question.
Bug reports and feature requests
GitHub Issues are reserved for bug reports, feature requests and other development-related subjects.
Related software
MuJoCo is the backbone for numerous environment packages. Below we list several bindings and converters.
Bindings
These packages give users of various languages access to MuJoCo functionality:
First-party bindings:
- Python bindings
- dm_control, Google DeepMind's related environment stack, includes PyMJCF, a module for procedural manipulation of MuJoCo models.
- JavaScript bindings and WebAssembly support (inspired stillonearth and zalo's community projects; mjswan extends these with real-time policy control, interactive force application, and more).
- C# bindings and Unity plug-in
Third-party bindings:
- MATLAB Simulink: Simulink Blockset for MuJoCo Simulator by Manoj Velmurugan.
- Swift: swift-mujoco
- Java: mujoco-java
- Julia: MuJoCo.jl
- Rust: MuJoCo-rs
Converters
- OpenSim: MyoConverter converts OpenSim models to MJCF.
- SDFormat: gz-mujoco is a two-way SDFormat <-> MJCF conversion tool.
- OBJ: obj2mjcf a script for converting composite OBJ files into a loadable MJCF model.
- onshape: Onshape to Robot Converts onshape CAD assemblies to MJCF.
Citation
If you use MuJoCo for published research, please cite:
@inproceedings{todorov2012mujoco,
title={MuJoCo: A physics engine for model-based control},
author={Todorov, Emanuel and Erez, Tom and Tassa, Yuval},
booktitle={2012 IEEE/RSJ International Conference on Intelligent Robots and Systems},
pages={5026--5033},
year={2012},
organization={IEEE},
doi={10.1109/IROS.2012.6386109}
}
License and Disclaimer
Copyright 2021 DeepMind Technologies Limited.
Box collision code (engine_collision_box.c)
is Copyright 2016 Svetoslav Kolev.
ReStructuredText documents, images, and videos in the doc directory are made
available under the terms of the Creative Commons Attribution 4.0 (CC BY 4.0)
license. You may obtain a copy of the License at
https://creativecommons.org/licenses/by/4.0/legalcode.
Source code is licensed under the Apache License, Version 2.0. You may obtain a copy of the License at https://www.apache.org/licenses/LICENSE-2.0.
This is not an officially supported Google product.