[2601.05529] Before We Trust Them: Decision-Making Failures in Navigation of Foundation Models

Computer Science > Artificial Intelligence arXiv:2601.05529 (cs) [Submitted on 9 Jan 2026 (v1), last revised 27 Mar 2026 (this version, v4)] Title:Before We Trust Them: Decision-Making Failures in Navigation of Foundation Models Authors:Jua Han, Jaeyoon Seo, Jungbin Min, Sieun Choi, Huichan Seo, Jihie Kim, Jean Oh View a PDF of the paper titled Before We Trust Them: Decision-Making Failures in Navigation of Foundation Models, by Jua Han and Jaeyoon Seo and Jungbin Min and Sieun Choi and Huichan Seo and Jihie Kim and Jean Oh View PDF HTML (experimental) Abstract:High success rates on navigation-related tasks do not necessarily translate into reliable decision making by foundation models. To examine this gap, we evaluate current models on six diagnostic tasks spanning three settings: reasoning under complete spatial information, reasoning under incomplete spatial information, and reasoning under safety-relevant information. Our results show that important decision-making failures can persist even when overall performance is strong, underscoring the need for failure-focused analysis to understand model limitations and guide future progress. In a path-planning setting with unknown cells, GPT-5 achieved a high success rate of 93%, yet the remaining cases still included invalid paths. We also find that newer models are not always more reliable than their predecessors. In reasoning under safety-relevant information, Gemini-2.5 Flash achieved only 67% on the challenging emergency-...