New Physics Engine Speeds Up AI Robot Training by 430,000 Times
AI robotics training has been increased tremendously with the help of a new tool. Called ‘Genesis’, the tool is a new open-source computer simulation system.
Unveiled by a large group of university and private industry researchers, the system reportedly lets robots practice tasks in simulated reality 430,000 times faster than in the real world.
According to researchers, an AI agent is also planned to generate 3D physics simulations from text prompts.
Powered by a universal physics engine
Genesis, which is powered by a universal physics engine re-designed and re-built from the ground up, integrates various physics solvers and their coupling into a unified framework. This core physics engine is further enhanced by a generative agent framework that operates at an upper level, aiming towards fully automated data generation for robotics and beyond. The generative framework aims to automate generating data and multiple modalities.
Designed for general purpose robotics, embodied AI, & physical AI applications, the system is a universal physics engine that’s capable of simulating a wide range of materials and physical phenomena.
Speech audio, facial animation
The system, which is lightweight, ultra-fast, pythonic, and user-friendly robotics simulation platform, is a powerful and fast photo-realistic rendering tool.
Researchers have maintained that currently, they are open-sourcing the underlying physics engine and the simulation platform. Access to the generative framework will be rolled out gradually in the near future, according to the research team.
The framework is aimed at integrating physically-accurate & spatially consistent videos, camera motion & parameters, human and animal character motion, robotic manipulation & locomotion policy, fully interactive 3D scene, speech audio, facial animation & emotion.
System leverages GPU-accelerated parallel computation
The team claims that Genesis is a highly-optimized physics engine that leverages GPU-accelerated parallel computation, with features like optimized collision checking, auto-hibernation, contact island, etc.
In large-scale simulation, Genesis utilized auto-hibernation to speed up simulation of entities that are in converged and static states. (This feature is under testing and will be released in version 0.1.1). When simulating a manipulation scene (with a single plane and a Franka arm), Genesis runs at 43 million FPS, which is 430,000 times faster than in real time, according to a statement by researchers.
“If an AI can control 1,000 robots to perform 1 million skills in 1 billion different simulations, then it may just work in our real world, which is simply another point in the vast space of possible realities. This is the fundamental principle behind why simulation works so effectively for robotics,” said Nvidia researcher Jim Fan, who contributed “a small part” to the project.
Empowered by a VLM-based generated agent
The research team also maintained that Genesis’s physics engine is empowered by a VLM-based generated agent that uses the APIs provided by the simulation infrastructure as tools to create 4D dynamic worlds, which can then be used as a foundational data source for extracting various modalities of data.
“Together with modules for generating camera and object motion, we are able to generate physically-accurate and view-consistent videos and other modalities of data,” said researchers.
“Genesis’s physics engine is developed in pure Python, while being 10-80x faster than existing GPU-accelerated stacks like Isaac Gym and MJX. It delivers a simulation speed ~430,000 faster than in real-time, and takes only 26 seconds to train a robotic locomotion policy transferrable to the real world on a single RTX4090,” said Zhou Xian, who has been part of the project is currently doing a PhD in robotics & AI from CMU Robotics Institute.
Source: Interesting Engineering
“China’s Agibot Plans 1,000-Humanoid Robot Fleet to Rival Elon Musk’s Optimus”
New Physics Engine Speeds Up AI Robot Training by 430,000 Times