AI coding agents can autonomously direct robot training

NVIDIA researchers, in collaboration with Carnegie Mellon University and the University of California, Berkeley, have developed an AI agent harness framework named ENPIRE that enables AI coding agents to autonomously direct robot training. In tests, these AI agents successfully taught robotic arms to perform complex tasks, including cutting zip ties and inserting GPUs into motherboard sockets, utilizing a "generous token budget" and available compute resources. The ENPIRE framework provides AI models with tools, memory, context management, constraint enforcement, and feedback loops, facilitating autonomous operation. Jim Fan, director of AI at NVIDIA, highlighted in a LinkedIn post that a portion of the NVIDIA GEAR lab now operates with this self-improvement capability overnight, with researchers reviewing the outcomes in the morning. This advancement demonstrates a significant step towards AI-driven automation in robot training and task execution.