AI coding agents taught robots how to install GPUs and cut zip ties
Ars Technica AI · Jeremy Hsu · 2026-06-17
NVIDIA GEAR lab and collaborators from CMU and UC Berkeley developed ENPIRE, an agentic harness framework that enables AI coding agents to autonomously design robot training programs, successfully teaching robotic arms to cut zip ties and install GPUs.
Extraction
Topics: roboticsai-agentsagentic-frameworksautonomous-robot-training
Claims
- NVIDIA GEAR lab developed ENPIRE, an agent harness framework that wraps AI models with memory, context, constraint, and feedback loop capabilities to enable autonomous robot training.
- AI coding agents using ENPIRE successfully trained robotic arms to perform dexterous tasks including cutting zip ties and inserting GPUs into motherboard sockets.
- The ENPIRE system operates autonomously overnight without human supervision, with researchers reviewing performance reports the following morning.
- ENPIRE was developed collaboratively by NVIDIA GEAR, Carnegie Mellon University, and UC Berkeley.
Key quotes
A part of our NVIDIA GEAR lab now self-improves tirelessly overnight. We just read the reports in the morning.
What happens when you give AI coding agents a lab full of robotic arms, some compute resources, and a 'generous token budget' for teaching the robots various tasks?