Reproduce “ARC-VLN: Adaptive Recursive Computing-based Visual-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

Xiang Zhang, Xiangyu Dong, Yang Lu, Xiaoguang Ma

ARC-VLN reduces storage costs in aerial navigation by sharing backbone weights and letting visual tokens exit early, while text and action tokens run to full depth. A uniform-target cross-entropy term helps prevent routing collapse. On OpenFly’s Test-Unseen subset, it improves navigation error, success rate, and SPL over OpenFly, but has lower oracle success rate; these are benchmark results, not flight validation.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.