Shulou(Shulou.com)11/24 Report--

CTOnews.com October 23 news, Nvidia recently announced an AI system called Eureka, which is based on OpenAI's GPT-4, allowing robots to perform more than 30 complex actions such as "pen turn,""drawer open,""scissors,""hands pass each other" and so on.

CTOnews.com learned after inquiry that NVIDIA Research led the development of Eureka, an AI system that allows developers to conduct reference learning with NVIDIA's own physics simulation software Isaac Gym.

Anima Anandkumar, senior director of Nvidia's AI research department, believes that although "reinforcement learning" has made progress over the past 10 years, there are still many challenges, such as "reward design" and other links, which are still in the "trial and error" stage. Nvidia's Eureka is designed to perform difficult tasks, combining generative AI and reinforcement learning algorithms for the first time.

Eureka's "reward design" scheme using GPT-4 can help robots learn trial-and-error and can take over 80% of human experts 'tasks, thus increasing the average robot training efficiency by more than 50%.

It is reported that when Eureka generates a "reward design" scheme, developers do not need to input additional task prompts or write predefined reward templates, combined with manual modification of rewards, to finally make robot actions more in line with developers 'intentions.

When the AI system cooperates with Issac Gym, it can use GPU acceleration to simulate, and quickly evaluate the quality of a large number of candidate feedback, so as to improve the training efficiency, and make statistical summary according to the training results, so as to improve the "reward scheme". While improving the response level of the robot, the AI system can be applied to a wider range of robot types, such as "animal type"(four feet),"human type"(two feet),"" quadcopter ", Robots such as arm models can be equipped with Eureka systems.

▲ Source Nvidia

▲ Image source Nvidia Nvidia conducted an "open source sensitivity benchmark test" on Eureka and obtained satisfactory results after evaluating the training results of 20 tasks.

Based on this, Nvidia believes that Eureka can achieve dexterous robot control and provide animators with a new way to "generate realistic physical movements."

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.