Before We Trust Them: Decision-Making Failures in Navigation of Foundation Models
Not on the map. Why?
Abstract (arXiv)
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 the current metrics may not capture critical limitations of the models and indicate good performance, 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 failed cases exhibit fundamental limitations of the models, e.g., the lack of structural spatial understanding essential for navigation. We also find that newer models are not always more reliable than their predecessors on this end. In reasoning under safety-relevant information, Gemini-2.5 Flash achieved only 67% on the challenging emergency-evacuation task, underperforming Gemini-2.0 Flash, which reached 100% under the same condition. Across all evaluations, models exhibited structural collapse, hallucinated reasoning, constraint violations, and unsafe decisions. These findings show that foundation models still exhibit substantial failures in navigation-related decision making and require fine-grained evaluation before they can be trusted.
Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 23 of its 35 references that are in the map. First posted on arXiv 9 Jan 2026.
Builds on
23 of 35 references.
References · 23
- 2024The Llama 3 Herd of Modelscited by 14,715
- 2023Improved Baselines with Visual Instruction Tuningcited by 4,418
- 2023RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Controlcited by 3,080
- 2025InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Modelscited by 1,834
- 2023Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!cited by 1,154
- 2023Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Promptingcited by 1,122
- 2023Towards Understanding Sycophancy in Language Modelscited by 1,040
- 2024SpatialVLM: Endowing Vision-Language Models with Spatial Reasoning Capabilitiescited by 706
- 2024Thinking in Space: How Multimodal Large Language Models See, Remember, and Recall Spacescited by 611
- 2020Beyond the Nav-Graph: Vision-and-Language Navigation in Continuous Environmentscited by 441
- 2020Room-Across-Room: Multilingual Vision-and-Language Navigation with Dense Spatiotemporal Groundingcited by 414
- 2024SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agentscited by 116
11 more
- 2019General Evaluation for Instruction Conditioned Navigation using Dynamic Time Warpingcited by 105
- 2024Sycophancy in Large Language Models: Causes and Mitigationscited by 101
- 2024OpenFMNav: Towards Open-Set Zero-Shot Object Navigation via Vision-Language Foundation Modelscited by 67
- 2023LangNav: Language as a Perceptual Representation for Navigationcited by 47
- 2025IS-Bench: Evaluating Interactive Safety of VLM-Driven Embodied Agents in Daily Household Taskscited by 41
- 2022Self-Prompting Large Language Models for Zero-Shot Open-Domain QAcited by 39
- 2025A Framework for Benchmarking and Aligning Task-Planning Safety in LLM-Based Embodied Agentscited by 30
- 2025SafeMERGE: Preserving Safety Alignment in Fine-Tuned Large Language Models via Selective Layer-Wise Model Mergingcited by 25
- 2024Evaluating Large Language Models on Spatial Tasks: A Multi-Task Benchmarking Studycited by 12
- 2025Decipher the Modality Gap in Multimodal Contrastive Learning: From Convergent Representations to Pairwise Alignmentcited by 10
- 2025FoREST: Frame of Reference Evaluation in Spatial Reasoning Taskscited by 9
Built on by
Not cited yet.