AI coding agents can autonomously direct robot training
Summary
Researchers developed a new software framework called ENPIRE that allows AI coding agents to independently train robots to perform complex tasks. Testing showed these AI agents could teach robots to do jobs like cutting zip ties and inserting GPUs with high success rates, sometimes better and faster than human-guided methods.Key Facts
- ENPIRE is a software system that helps AI agents control robot training by managing tasks like resetting robots, checking results, and improving training strategies.
- The framework was created by NVIDIA’s GEAR lab with help from Carnegie Mellon University and UC Berkeley.
- AI agents tested include OpenAI’s Codex with GPT-5.5, Anthropic’s Claude Code with Opus 4.7, and Moonshot AI’s Kimi Code with Kimi K2.6.
- These AI agents independently developed and tested different ways to train robots, keeping the best improvements.
- Tasks mastered include moving T-shaped blocks, organizing pins, cutting zip ties, and inserting GPUs into motherboards.
- Teams of eight AI agents trained robots faster and more successfully than smaller teams or single agents.
- The AI groups achieved nearly 100% success in some tasks faster than human-guided methods.
- Limitations include robots being idle while AI agents are busy debugging or waiting for processing time.
Read the Full Article
This is a fact-based summary from The Actual News. Click below to read the complete story directly from the original source.