On Thursday, researchers introduced Genesis, a groundbreaking generative artificial intelligence (AI) physics model capable of simulating four-dimensional (4D) worlds. This innovative model distinguishes itself by integrating various functionalities to simulate scenarios tailored for general robotics and physical AI applications. The team behind Genesis asserts that it achieves remarkable simulation speeds, operating up to 80 times quicker than conventional GPU-accelerated systems. The open-source AI model is accessible through the Python Package Index (PyPI); however, users will also need to install PyTorch to utilize it.
Genesis AI Physics Model Can Simulate Dynamic Worlds for Robotics Training
Zhou Xian, the lead researcher on the project, announced the launch of Genesis in a post on X (formerly Twitter). He noted that the project emerged from a two-year collaborative effort among over 20 research labs, resulting in a sophisticated integration of multiple physics solvers into a cohesive framework.
Genesis supports simulating various types of physical phenomena. We developed from scratch a unified physics engine that integrates various SOTA physics solvers (MPM, SPH, FEM, Rigid Body, PBD, etc.), supporting simulation of a wide range of materials: rigid body, articulated… pic.twitter.com/PqhIWULKgp
— Zhou Xian (@zhou_xian_) December 18, 2024
Constructed entirely with Python, Genesis features a generative agent framework supported by a universal physics engine. Currently, the team has released the foundational physics engine and simulation platform as open source, with plans to unveil the generative framework at a future date.
The potential of Genesis is substantial, based on researchers’ claims. It is reportedly between 10 and 80 times faster than similar systems like Isaac Gym and MJX, which depend on GPU acceleration for simulations. In some instances, the engine is said to achieve simulation speeds that are over 430,000 times faster than real-time. The researchers added that it can effectively train a robotic locomotion policy using a single Nvidia RTX4090 GPU in just 26 seconds.
Among the notable features of Genesis is its comprehensive integration with Python, serving as both the frontend and backend framework. The system also offers an application programming interface (API) for added accessibility. While boasting impressive simulation speeds, it is designed to retain accuracy and fidelity. The unified framework accommodates multiple physics solvers, facilitating the simulation of various physical phenomena and materials. Furthermore, the physics engine includes ray-tracing rendering capabilities.
