Nvidia Unveils ENPIRE Open-Source Agentic Robots

Nvidia's ENPIRE project uses AI agents to teach robots high-precision tasks like GPU installation through real-world self-improvement.

Nvidia Director of AI Jim Fan showcased the ENPIRE project, an open-source initiative detailed in the research paper "ENPIRE: Agentic Robot Policy Self-Improvement in the Real World." The system utilized eight Codex agents equipped with a fleet of robots, GPU allocations, and a token budget to practice skills directly on hardware. During demonstrations, robots performed high-precision tasks including sorting metal pins, cutting zipties, and installing a graphics card into a motherboard, which involved one robot arm passing the component to another. Researchers tested various models including Codex with GPT-5.5, Claude Code with Opus 4.7, and Kimi Code with Kimi K2.6. The study concluded that eight robots exploring in parallel solve tasks faster than smaller groups, allowing the machines to tinker with the control stack to master complex real-world actions.