Nvidia has built a system in which robots help train themselves, with AI coding agents directing the work instead of human researchers, according to reporting from Ars Technica, Decrypt, and Tech Times.
Ars Technica describes the effort as a "self-improvement program for robots" that enlists teams of AI coding agents. Rather than having engineers hand-write and tune the software that teaches a robot a new skill, these agents can autonomously direct the training process themselves.
The project is named ENPIRE, according to Tech Times, which reports that the system "closes the loop" by letting AI agents run robotics research on real, physical hardware — not just in simulation. Decrypt frames the result plainly: Nvidia has "built robots that train themselves using AI coding agents."
In practical terms, the sources describe a feedback loop. AI agents write and adjust the code that governs how a robot learns, the robot attempts tasks on actual hardware, and the results feed back into the next round of agent-directed adjustments. That removes much of the slow, manual trial-and-error that has long bottlenecked robotics work.
The sources do not specify which robots, tasks, or performance figures are involved, so the real-world capabilities remain to be demonstrated beyond Nvidia's own description.
Why it matters: if AI agents can reliably run their own robotics experiments on real machines, the pace of robot development could shift from being limited by human engineering hours to being limited mainly by available computing power — a change that would accelerate how quickly capable robots reach factories, warehouses, and eventually everyday life.