University of Pennsylvania researchers develop SymSkill for rapid robot learning

Researchers at the University of Pennsylvania have developed SymSkill, a system that enables robots to master complex, multi-step tasks using only five minutes of video training. The system combines symbolic planning to determine task sequences with a physics-based dynamical systems approach to control physical movement.

This dual-level approach allows robots to adapt to environmental changes, such as human interference or failed attempts, without needing to be trained on every possible variation of a task. During testing, the robot successfully performed sequences like removing a lid and moving a banana from a pan to a plate, adjusting its actions in real-time when interrupted.