Accelerating physics computations with Python

Company Releases Open-Source Python Library for Fast Robot Simulation with MuJoCo

After a full year of hands-on robotics research, our team is sharing one of the tools that made that work possible: a high-performance Python library built on top of the MuJoCo physics engine. Starting today, the library is available as an open-source project, free for anyone to use, study, and improve.

Why we built it

Robotics research moves fast, and researchers often spend a surprising amount of time writing the same low-level code over and over: setting up simulation scenes, stepping the physics forward, handling sensors and actuators, and wiring everything into learning pipelines. We ran into those same pain points repeatedly during our own projects.

Rather than keep patching together one-off scripts, we consolidated the patterns that worked best into a single library. The goal was simple: make robot simulation in Python faster to write and faster to run, without hiding the flexibility that researchers need when they want to customize every detail.

What the library does

At its core, the library is a wrapper and toolkit around MuJoCo, a widely used physics engine known for accurate and efficient simulation of articulated bodies — everything from robot arms to legged robots. On top of that foundation, the library adds:

• Clean Python APIs for building simulation environments, so you can describe a robot and a task in a few lines of code instead of hundreds.
• Performance-focused design, tuned during a year of daily research use, to keep simulation loops fast even in large-scale experiments.
• Compatibility with common research workflows, making it straightforward to plug simulated environments into training or evaluation pipelines.

Because it is open source, everything is transparent: you can read the code, modify it for your hardware or task, and contribute fixes or features back to the project.

Who it is for

The library is aimed primarily at robotics researchers and engineers who already work in Python and want a dependable simulation layer without building one from scratch. It should also be useful to students and hobbyists who want to experiment with robot physics using a tool that scales beyond toy examples.

If your work depends heavily on simulating robots — whether for control algorithms, learning-based methods, or benchmarking new designs — this library is designed to save you setup time and give you a solid, tested foundation.

What happens next

Open-sourcing the library is the beginning, not the end. As more people use it, we expect bugs to surface, features to be requested, and use cases we never anticipated to appear. Community feedback and contributions will shape how the project evolves, and we plan to keep maintaining it as part of our ongoing robotics work.

The code, documentation, and examples are available now in the project repository. We are excited to see what the robotics community builds with it — and we hope it removes at least a few of the hurdles we spent our own year wrestling with.