Meet Genesis, an Open-Source Universal Physics Engine That Generates 4D Worlds

Zhou Xian, Yiling Qiao, Zhenjia Xu, Tsun-Hsuan Wang, and other core contributors from institutions including Carnegie Mellon University, Stanford University, MIT (CSAIL), NVIDIA, and Tsinghua University have introduced Genesis, a new physics platform to support a range of robotics and embodied AI applications.

“Announcing the Genesis project — after a 24-month large-scale research collaboration involving over 20 research labs — a generative physics engine able to generate 4D dynamical worlds, powered by a physics simulation platform designed for general-purpose robotics and physical AI applications,” said Zhou Xian, one of the researchers on the project.

Genesis is described as “the world’s fastest physics engine,” offering simulation speeds up to 80 times faster than existing GPU-accelerated robotic simulators, such as Isaac Gym and Mujoco MJX , without sacrificing accuracy. 

The platform enables the generation of 4D dynamic worlds, providing a foundation for data extraction. Genesis also supports character motion generation, including detailed animations such as a miniature Wukong performing acrobatic movements, and robotic policy generation for a range of scenarios, from robotic arms organizing books to drones performing synchronized flips.

Genesis is compatible with differentiable simulation, supporting a wide range of solvers, including MPM and Tool Solvers. The platform’s key features include “100% Python” integration, an “extremely simple and user-friendly API design,” and “photo-realistic ray-tracing rendering.”

 Additionally, Genesis supports generative simulation, enabling users to generate data from natural language descriptions for tasks such as scene creation, motion generation, and video simulation.

The platform integrates various physics solvers into a unified framework designed to simulate materials and physical phenomena with high fidelity. Built entirely in Python, Genesis provides a fast afnd user-friendly simulation environment aimed at transforming data generation in robotics.

Genesis is optimised for speed, with GPU-accelerated parallel computation to achieve high simulation performance. In one example, a manipulation scene using a Franka arm runs at 43 millision FPS, 430,000 times faster than real time. The platform also employs auto-hibernation for efficient simulation of static entities, with a planned feature release in version 0.1.1.

Genesis physics engine and simulation are open-source, with the generative framework expected to be released soon. “Our mission is to lower the barrier to using physics simulations and make robotics research accessible to everyone,” the blog stated.

The platform can be installed via PyPI with simple commands, and the full documentation and API reference are available for users. The Genesis project is open to contributions from the community to further enhance its capabilities.