This humanoid robot learned realistic lip movements by watching YouTube

Twenty-six miniaturized motors beneath synthetic skin have helped a humanoid robot at Columbia University cross a crucial threshold: learning to move its lips like a human speaker. The breakthrough came not through programming every phoneme, but by letting the robot watch itself make thousands of random expressions in a mirror before studying hours of YouTube videos to understand how human lips sync with sound. "The more it interacts with humans, the better it will get," said Hod Lipson, director of the Creative Machines Lab, whose team created the first autonomous system to acquire natural lip movement through visual learning alone.

The achievement tackles robotics' persistent uncanny valley problem, where stiff, puppet-like facial movements shatter the illusion of life. While the robot still struggles with hard consonants and lip-puckered sounds, it can now synchronize speech across multiple languages and even sing from its AI-generated album "Hello World." As billions of humanoid robots may enter production within a decade, researchers believe believable facial expressions will prove as crucial as functional mobility for machines working in healthcare, education, and elder care. "We are close to crossing the uncanny valley," said study lead Yuhang Hu. "There is no future where humanoid robots don't have faces that move properly."