ASPIRE: Agentic Skill Discovery for Robotics

ASPIRE is a continual-learning system for robotics that autonomously writes, executes, diagnoses, repairs, and validates robot-control programs in a code-as-policy framework. Validated fixes are distilled into a reusable skill library, while evolutionary search creates diverse task sequences and programs beyond single-trajectory refinement. The paper reports gains of up to 77% on LIBERO-Pro manipulation under perturbation, 72% on Robosuite bimanual handover, and 32% on BEHAVIOR-1K long-horizon household tasks. On LIBERO-Pro Long, ASPIRE reaches 31% success versus 4% for prior methods, and reports initial evidence that simulation-discovered skills can reduce real-robot programming effort across embodiments and robot APIs.