TactAlign: Human-to-Robot Policy Transfer via Tactile Alignment

Researchers from UC Berkeley and Meta have developed TactAlign, a new method that transfers skills from humans to robots using only tactile feedback. It transforms touch signals from a human wearing a tactile glove into a shared language a robot can understand, without needing perfectly matched data. The system enables robots to learn contact-rich tasks like pivoting, insertion, and lid closing in under 5 minutes of human demo data and can even perform zero-shot transfer for highly dexterous tasks like screwing in a lightbulb.