NVIDIA AI Introduces ASPIRE: A Self-Improving Robotics Framework Reaching 31% Zero-Shot on LIBERO-Pro Long Tasks

ASPIRE is a continual learning system that enables robots to write and refine their own control programs while building a reusable library of validated fixes. Traditional robot programming requires manually handling perception, motion planning, grasping, and failure cases, but code-as-policy systems let language models compose these into executable programs that can be inspected and debugged. ASPIRE improves upon existing coding agents by providing detailed per-primitive feedback about what specifically failed during execution, rather than just coarse task-level signals, and it preserves solutions as transferable skills instead of discarding them after each task. In benchmarks, ASPIRE achieved 72% on LIBERO-Pro overall and notably reached 31% on held-out long-horizon tasks without prior training on them, compared to about 4% for previous methods.