Reproduce “StopWatch: Goal Grounding for Object-goal 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

Xinhao Liu, Yunsheng Tian, Yuheng Qiu, Pieter Abbeel, Xi Chen, Guanya Shi, Chen Feng

StopWatch separates perception and policy models for open-vocabulary object-goal navigation. A perception model predicts visible object categories and goal distance from RGB, while a separately trained policy uses those cues with observation history to choose actions. Scene annotations supervise grounding at every demonstration frame. On clean HM3D-OVON validation-unseen, StopWatch reaches 69.6% success versus 60.8% for LightNav-0 in simulation.

Summary by alphaXiv.

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