Nvidia has announced the development of a new AI agent called Eureka, based on OpenAI's GPT-4 model, capable of autonomously teaching robots complex skills.
“Eureka independently wrote the reward algorithm and, for the first time, taught a robotic hand to perform rapid pen-twisting techniques,” the company said in a blog post. “It also taught the robot how to open drawers and cabinets, throw and catch balls, and use scissors, among about 30 other tasks.”
Reinforcement learning has achieved amazing results over the past decade, but many challenges remain, such as designing rewards, which remains a process of trial and error.
Eureka is a first step in developing new algorithms that can combine generative and reinforcement learning methods to solve difficult tasks.
Nvidia also released the Eureka AI Algorithm Library so users can try out Nvidia Isaac Gym, a benchmark physics simulation implementation for reinforcement learning research.
Nvidia built Nvidia Isaac Gym on Nvidia Omniverse, a development platform for creating 3D tools and applications based on the OpenUSD framework.
There has been a lot of talk about AI agents in recent months, and discussions intensified in April with the emergence of standalone AI agents such as Auto-GPT, BabyAGI, and AgentGPT.
Nvidia's current work builds on previous work, including Voyager, a GPT-4-based AI agent that can run Minecraft autonomously.
“This is a huge business opportunity that could be worth billions of dollars,” Jeff Clune, a computer science professor at the University of British Columbia and a former OpenAI researcher, wrote in a New York Times article about chatbots this week. . He added that this has a great positive impact on society.
“Eureka leverages ZSL's next-generation machine learning methods, code typing, and context-aware optimization capabilities of large complex language models such as GPT-4 to scale reward code,” the authors said in a new paper. Can be used to learn complex skills. » Thanks to reinforcement learning.
“Eureka is a unique combination of large language modeling and GPU-accelerated simulation technology,” NVIDIA said. “We believe Eureka enables intelligent control of robots and provides artists with a new way to create realistic animations.”