Reproduce “AeroRSI: Recursive Self-Improvement for Aerial Vision-Language Navigation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shiwei Wu, Yuwen Pan, Zhi Li, Yaowen Li, Yan Zhou, Xuemiao Xu, Yu Liu
AeroRSI improves a supervised 2B-parameter aerial vision-language-action policy by testing candidate repair actions through continued simulator flights, then repeating training at failures exposed by the updated policy. On TravelUAV, success rate rises from 46.39% to 67.49% after three rounds, versus 56.87% for AeroDPO from the same initialization; results are simulator-specific, and the final-round gain is within evaluation noise.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.