Aux-Think: Exploring Reasoning Strategies for Data-Efficient Vision-Language Navigation

2025-05-17

AI Summary

Visual-language navigation (VLN) suffers from poorly understood reasoning strategies and frequent reasoning collapse during inference under data-efficient settings. Method: We propose Aux-Think—a novel framework that internalizes structured, chain-of-thought (CoT)-guided reasoning via supervised multimodal training, and decouples “reasoning” from “action” at inference time to directly output navigation decisions. Contributions/Results: (1) We systematically identify and characterize the reasoning collapse phenomenon in VLN for the first time; (2) we introduce a training–inference decoupling paradigm; and (3) we construct R2R-CoT-320k—the first large-scale, CoT-annotated dataset for VLN. Experiments demonstrate that Aux-Think achieves significantly higher navigation accuracy and reduced training cost compared to baselines under equivalent data budgets, empirically validating the superiority of internalized reasoning over online reasoning.

Technology Category

• Computer Vision: Visual Reasoning & Symbolic Representations
• Natural Language Processing: Safety and Robustness
• Planning, Routing, and Scheduling: Model-Based Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Vision-Language Navigation (VLN) is a critical task for developing embodied agents that can follow natural language instructions to navigate in complex real-world environments. Recent advances in VLN by large pretrained models have significantly improved generalization and instruction grounding compared to traditional approaches. However, the role of reasoning strategies in navigation-an action-centric, long-horizon task-remains underexplored, despite Chain-of-Thought (CoT) reasoning's demonstrated success in static tasks like visual question answering. To address this gap, we conduct the first systematic evaluation of reasoning strategies for VLN, including No-Think (direct action prediction), Pre-Think (reason before action), and Post-Think (reason after action). Surprisingly, our findings reveal the Inference-time Reasoning Collapse issue, where inference-time reasoning degrades navigation accuracy, highlighting the challenges of integrating reasoning into VLN. Based on this insight, we propose Aux-Think, a framework that trains models to internalize structured reasoning patterns through CoT supervision, while inferring action directly without reasoning in online prediction. To support this framework, we release R2R-CoT-320k, the first Chain-of-Thought annotated dataset for VLN. Extensive experiments show that Aux-Think reduces training effort greatly and achieves the best performance under the same data scale.

Problem

• Evaluating reasoning strategies for Vision-Language Navigation (VLN)
• Addressing Inference-time Reasoning Collapse in VLN tasks
• Proposing Aux-Think to internalize reasoning without online prediction

Innovation

• Evaluates reasoning strategies for Vision-Language Navigation
• Proposes Aux-Think framework with CoT supervision
• Introduces R2R-CoT-320k dataset for VLN