Updated
Updated · Ars Technica · Jun 17
AI Coding Agents Train Robots to Insert GPUs Using 4-Module ENPIRE Harness
Updated
Updated · Ars Technica · Jun 17

AI Coding Agents Train Robots to Insert GPUs Using 4-Module ENPIRE Harness

3 articles · Updated · Ars Technica · Jun 17

Summary

  • Nvidia GEAR researchers said AI coding agents autonomously built and refined robot-training regimens that taught robotic arms to cut zip ties and insert GPUs into narrow motherboard sockets.
  • ENPIRE enabled that loop with four modules for task reset and verification, policy refinement, parallel evaluation across multiple physical robots, and failure analysis that fed back into code, infrastructure, and algorithm changes.
  • Three coding-agent systems—OpenAI Codex with GPT-5.5, Anthropic Claude Code with Opus 4.7, and Moonshot Kimi Code with Kimi K2.6—independently tried different approaches and kept changes that improved success rates over repeated real-world tests.
  • A June 16 paper details the framework, and Nvidia says it will open-source the setup so others can run similar self-directed robot labs beyond the company’s own overnight research workflow.

Insights

NVIDIA's robots now train themselves overnight. Is the promise of a home-based AI research lab finally within reach?
If an AI can endlessly improve itself, what guardrails can truly keep humanity in control of its evolution?
As AI learns to automate automation itself, what uniquely human skills will remain valuable in our economy?