Reproduce “EvoNav-Bench: Benchmarking Lifelong Navigation in Evolving Environments” 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

Xilin Wang, Guoxi Zhang, Hongming Xu, Zhuofan Zhang, Tianxu Wang, Lifeng Fan

EvoNav-Bench tests lifelong navigation when rooms change between tasks without warning. It uses simulated object relocations in ProcTHOR homes and compares agents under ground-truth perception. In a controlled Stage-2 comparison, 3D-Mem success falls from 92.2% in unchanged scenes to 40.0% after change, while Stage-Reset shifts from 60.0% to 57.8%. This suggests retained entries can mislead, but does not prove they are the only cause.

Summary by alphaXiv.

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