Imagine teaching a robot to maneuver in an environment without ever letting it see the real world. That’s what researchers at the Massachusetts Institute of Technology have done. Engineers combined generative AI with physics simulators to enable a quadruped robot to learn from synthetic data rather than real-world experience. Robots trained using machine-generated environments succeeded 88 percent of the time, while those taught by people managed only 15 percent. The goal is to help robots adapt more quickly to new tasks and environments. That will be important for applications such as search and rescue.

This is an excerpt from our April episode of Assembly News Now. Watch the full episode here!
 

Don't miss an episode!