Robotic software learns skills across different hardware

Researchers at EPFL have developed a new framework called Kinematic Intelligence, enabling robots to transfer learned skills between different hardware configurations, much like syncing apps between smartphones. This innovation addresses the long-standing challenge of retraining robots from scratch when switching models due to varying designs and physical constraints. The system embeds a deep mathematical awareness of the robot's physical limitations directly into its control policy, avoiding the need for AI-based solutions or post-training corrections. This approach ensures that demonstrated skills can be executed safely and reliably by entirely different robot types, overcoming issues like singularities where control can be lost.