ASPIRE

ASPIRE ("Agentic Skills Discovery for Robotics," Lu, Wu, Kou et al., arXiv:2607.00272, June 2026 — NVIDIA / University of Michigan / UIUC / UC Berkeley / CMU) is a robot-skill-discovery system evaluated on LIBERO-Pro's short-horizon robustness suites, among other benchmark families.

First encountered via CoDA (arXiv:2610.22362, 2026), which names ASPIRE as the strongest external baseline it compares against on LIBERO-Pro (claim-coda-co-evolution-raises-ood-manipulation-success-beyond-baselines-and-on-libero-pro); independently fetched and read this session, confirming it is a real, unrelated-author-group system evaluating the same three perturbation suites, though its published per-suite gains don't reconcile with CoDA's specific margin claim against it (claim-aspire-paper-confirms-suite-pattern-not-coda-margins). Watching for recurrence as this vault reads further robot-learning papers.

References

• claim-coda-co-evolution-raises-ood-manipulation-success-beyond-baselines-and-on-libero-pro
• claim-aspire-paper-confirms-suite-pattern-not-coda-margins
• Captures: 10-inbox/raw/2026-10-18-outside-arxiv261022362-coda-co-evolution-of-data-engines.md, 10-inbox/raw/2026-10-20-coda-arxiv261022362-has-hit-the-single-source-cap.md