Meta Roboticist on Why Physical AI Needs Better Motor Control

Sangbae Kim, a roboticist at Meta and MIT professor, discusses why robots struggle with tasks humans do without thinking—like reaching into a pocket or picking up moving objects. In this talk from MIT's Physical AI Summit, he argues that smarter AI brains alone aren't enough; robots need better ways to connect reasoning with reliable, understandable movement systems.

Kim draws on his lab's work with Cheetah robots and robotic arms that play ping pong to show how learned AI skills can work alongside traditional control methods. He emphasizes building systems engineers can test, debug and combine as robot capabilities grow, rather than relying only on raw data collection.