文献类型：期刊文献

英文题名：Dual-modal chain-of-thought reasoning for continuous vision-language navigation

作者：Liu, Man[1];Wu, Danfeng[2];Xu, Yaru[2];Zhou, Fenfen[2];Zhang, Xueyan[2]

第一作者：Liu, Man

通讯作者：Liu, M[1];Wu, DF[2]

机构：[1]Liaoning Tech Univ, Coll Software, 188 Longwan South St, Huludao 125105, Liaoning, Peoples R China;[2]Beijing Union Univ, Coll Robot, 4 Gongti North Rd, Beijing 100027, Peoples R China

第一机构：Liaoning Tech Univ, Coll Software, 188 Longwan South St, Huludao 125105, Liaoning, Peoples R China

通讯机构：[1]corresponding author), Liaoning Tech Univ, Coll Software, 188 Longwan South St, Huludao 125105, Liaoning, Peoples R China;[2]corresponding author), Beijing Union Univ, Coll Robot, 4 Gongti North Rd, Beijing 100027, Peoples R China.|[1141739]北京联合大学机器人学院;[11417]北京联合大学;

年份：2026

卷号：206

外文期刊名：ROBOTICS AND AUTONOMOUS SYSTEMS

收录：;EI(收录号：20263421354543);Scopus(收录号：2-s2.0-105047909237);WOS:【SCI-EXPANDED(收录号:WOS:001855651800001)】；

基金：This work was supported by Beijing Union University's 2026 Key Major Cultivation Project (No. ZK20260203); Beijing Union University's 2026 High-level Comprehensive Course Construction Project; and Beijing Union University's 2026 School-level Virtual Simulation Experimental Teaching Project (No. JJ2026SYP001).

语种：英文

外文关键词：Vision-language navigation; Chain-of-thought; Large Vision-Language Model; Dual-modal action; Reinforcement learning

摘要：Embodied intelligence requires an agent to complete autonomous navigation in continuous environments following natural language instructions. Existing methods typically output mid-level natural language actions or low-level control commands. However, there is a lack of effective coordination between high-level reasoning and continuous control, which leaves room for improvement in decision interpretability and control stability. To address these issues, we propose the VLN-COT framework, which integrates vision-language navigation models with explicit chain-of-thought reasoning. This framework introduces a generative reasoning mechanism to enhance decision interpretability through structured reasoning processes. It also adopts a dual-modal action representation that simultaneously predicts mid-level natural language actions and low-level continuous control commands, achieving joint optimization between high-level reasoning and low-level execution. During training, we combine supervised fine-tuning with GRPO-based reinforcement learning and design three complementary reward functions to obtain a stable and interpretable navigation decisions. Experimental results on the VLN-CE benchmark show that VLN-COT achieves significant improvements in key metrics such as success rate and path efficiency, validating the effectiveness of explicit reasoning and dual-modal action representation in continuous vision-language navigation.