Reproduce “Data-Centric Recursive Self-Improvement for Navigation via Policy-Synchronized Language-Action Dual Bootstrapping” 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
Songze Li, Haoming Gu, Ziyang Gong, Zhitong Wang, Gengze Zhou, Zun Wang, Yuhan Zhu, Tianxiang Jiang and 6 more
RSI-Nav trains a shared model to generate instructions for unlabeled routes and follow them. It retains pairs only when execution reaches the target within 3 meters efficiently, then updates before the next trajectory shard. On R2R-CE validation-unseen, success rate rises from 52.9% at supervised cold start to 68.6% after recursive evolution. Execution does not establish semantic accuracy or real-world robustness.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.