ASPIRE: Robots Learn and Share Skills Continuously

A new system called ASPIRE has been developed to allow robots to build and continuously evolve a library of skills. This means that a robot's ability to perform a task improves over time, rather than starting from scratch with each new challenge. The system works by having coding agents observe various sensory inputs from both simulated and real-world robot interactions, then using an evolutionary search to refine control programs and store the most effective strategies in a growing knowledge base. ASPIRE redefines continual learning for robotics, focusing on skill refinement rather than traditional gradient descent. It addresses the notoriously difficult 'sim-to-real' gap by transferring learned 'know-how' or strategies, rather than raw visual data or model weights. This allows robots to adapt to real-world environments much faster, reducing the need for extensive retraining. The project, a collaboration between NVIDIA GEAR lab, UMich, Berkeley, and CMU, promises to open-source its full stack, enabling other robotic systems to benefit from its expanding skill library.