XNews

ENPIRE: 8 Codex Agents Autonomously Master Robot Skills via Physical World Research

ENPIRE: 8 Codex Agents Autonomously Master Robot Skills via Physical World Research

ENPIRE gives 8 Codex agents a fleet of robots, GPUs, and token budget, tasking them to autonomously solve physical tasks as quickly and safely as possible. Without human intervention, the agents look for visual clues, reset scenes, practice skills, tinker with the control stack, read research papers online, debate, get stuck, and retry directly on hardware. The system solves high-precision tasks such as tying zip-ties, organizing fine pins, and installing GPUs. It exhibits physical scaling: 8 robots exploring in parallel improve significantly faster than fewer. NVIDIA's GEAR lab now self-improves overnight, and the full stack will be open-sourced.

Source posts