# OpenAI Robots Learn to Adapt in Real Time After Training Only in Virtual Worlds

OpenAI announced a significant breakthrough in robotics: their AI systems can now train entirely in simulated environments and then adapt to unexpected real-world changes when deployed on physical robots.

The key advancement is the shift from "open-loop" to "closed-loop" systems. Previously, robots trained in simulation would execute pre-programmed sequences without adjusting to their surroundings. The new approach allows robots to continuously monitor their environment and react to unplanned changes as they work—similar to how humans adjust their grip if an object starts to slip.

This matters because training robots in the physical world is expensive, time-consuming, and can damage equipment. By perfecting skills in simulation first, developers can run thousands of training scenarios quickly and safely. The challenge has always been transferring those simulated skills to messy, unpredictable reality—a problem