Nvidia Releases GPT-Powered Robot Training Software

Nvidia has released a new AI-enabled system, the company says can teach robots complex skills.

The Eureka system, which is powered by GPT-4, autonomously creates reward algorithms to train robots. The software is also self-learning, generating insights from each robot training session and modifying its algorithm generation in response.

The company demonstrated the system by training a robotic hand to spin a pen between its fingers, demonstrating its dexterity.

The pen-spinning capabilities are one of nearly 30 new tasks robots can be taught using Eureka, including opening drawers, catching and throwing a ball, and using scissors.

According to a paper released by Nvidia as part of the announcement, the reinforcement-learning system allows robots to outperform human-written algorithms on more than 80% of the tasks considered. 

“Reinforcement learning has enabled impressive wins over the last decade, yet many challenges still exist, such as reward design, which remains a trial-and-error process,” said Anima Anandkumar, Nvidia’s senior director of AI research. “Eureka is a first step toward developing new algorithms that integrate generative and reinforcement learning methods to solve hard tasks.”

“Eureka is a unique combination of large language models and Nvidia…simulation technologies,” said Linxi ‘Jim’ Fan, Nvidia’s senior research scientist. “We believe that Eureka will enable dexterous robot control and provide a new way to produce physically realistic animations for artists.”

Related:Nvidia, Segway-Ninebot Launch Robotics Development Platform