Reproduce “NavGPT-3: Harnessing Context in a Hierarchical Navigation Runtime” 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
Gengze Zhou, Yicong Hong, Jiazhao Zhang, Xunyi Zhao, Jian Zhou, Zixing Lei, Zun Wang, Chongyang Zhao and 4 more
NavGPT-3 combines a reasoning Planner, learned action policy, spatial tools, and runtime to coordinate route review and motion. The policy predicts eight waypoints and executes only the first before predicting again. On full RxR-CE Val-Unseen, success rises from 82.39 to 90.43, with 78.47 nDTW. Human scores are 90.4 SR and 77.7 nDTW, but use discrete graph edges rather than continuous motion.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.