AI Coding Agents Autonomously Train Robots to Install GPUs and Cut Zip Ties

Researchers at NVIDIA’s GEAR lab, in collaboration with Carnegie Mellon University and UC Berkeley, developed a framework called ENPIRE that enables AI coding agents to autonomously train robots. Using robotic arms, compute resources, and a generous token budget, the agents devised training regimens that taught robots to cut zip ties and insert GPUs into motherboard sockets. According to NVIDIA AI director Jim Fan, part of the lab now self-improves tirelessly overnight, with researchers reviewing results each morning.