Seonglae ChoAgent Skills Discovery Methods
ASPIRE: Agentic Skills Discovery for AI Robotics
NVIDIA's ASPIRE treats continual learning as iterative refinement of sensorimotor skills stored in a repository rather than gradient updates to model weights. The v1.0 README reports 99% success across 80 LIBERO-Pro tasks and selected Robosuite and BEHAVIOR-1K tasks, and presents separate real-world demonstrations including Rubik's Cube solving.