The team led by Runyu Lu and colleagues introduced ASPIRE, a system that writes robot control code, detects failures, repairs the code and adds the fixes to a growing skill library. In simulation ASPIRE beats earlier methods by up to 77% on perturbed manipulation tasks and reaches 99% success on the 80 LIBERO‑Pro benchmarks when paired with Claude Opus 5.5. The same library can be reused across different robots and even helped a real robot solve a Rubik’s Cube more efficiently.
Why it matters
For robot engineers, ASPIRE could cut the time spent hand‑coding tasks and make it easier to adapt robots to new jobs.
