NVIDIA has developed 'ENPIRE,' a system that uses AI agents such as Claude Code to autonomously improve robots.
NVIDIA has announced ENPIRE , a harness framework that brings the autonomous capabilities of AI agents to robots. Developed in collaboration with researchers from Carnegie Mellon University and the University of California, Berkeley, ENPIRE can enhance robots' task-performing abilities using AI agents such as Claude Code and Codex.
ENPIRE: Agentic Robot Policy Self-Improvement in the Real World
Creating dexterous robots capable of performing intricate tasks requires 'human supervision' and 'human algorithm development.' However, NVIDIA points out that human supervision and development are bottlenecks in developing general-purpose physical intelligence.
ENPIRE is a harness framework for applying existing AI agents such as 'Claude Code,' 'Codex,' and 'Kimi Code' to robot development. While AI agents can autonomously develop algorithms for robots in a digital environment, they cannot apply them to real-world robots. Therefore, NVIDIA developed 'ENPIRE' as a harness framework that allows algorithms to be applied to real-world robots and the results to be verified, enabling the application of AI agent autonomy to robot development. The name ENPIRE comes from ' Environment ,' ' Policy Improvement,' ' Rollout ,' and 'Evolution.'
By using ENPIRE, you can continuously improve the robot's operation by executing a loop of 'executing tasks,' 'planning improvements based on results,' 'implementing improvements into the algorithm,' and 'applying the improved algorithm to the robot.' NVIDIA's experiments have shown that it can autonomously improve the success rate of actions such as 'inserting pins,' 'setting GPUs onto circuit boards,' and 'cutting cable ties,' achieving a success rate of 99%.
ENPIRE also has the advantage that 'increasing the number of robots operated at once improves the speed of task improvement.' However, ENPIRE also has problems such as 'the AI agent not making full use of the robot's resources, resulting in wasted GPUs' and 'increasing the number of robots operated at once increases token consumption and thus increases costs.'
Related Posts:
in AI, Posted by log1o_hf