NVIDIA ENPIRE Framework Lets AI Coding Agents Autonomously Train Eight-Robot Fleet to 99% Precision Task Success With No Human Intervention

NVIDIA's ENPIRE framework, released June 17 by NVIDIA GEAR Lab in collaboration with CMU and UC Berkeley, enables AI coding agents—Codex, Claude Code, and Kimi Code—to autonomously run the complete robot training cycle on physical hardware: resetting scenes, executing trials, analyzing failures, revising training code, and iterating until policies converge, with no human involvement between steps. An eight-robot fleet reached 99% success on precision tasks including pin insertion, GPU installation, and zip-tie cutting; scaling from one to eight robots cut mastery time by over 50% while token costs scaled faster than time savings. The system is the first to demonstrate that AI agents can conduct iterative experimental science directly on physical robot hardware rather than in simulation.