Research map / Vision-language and multimodal / Robot action and navigation
NavCoT: Boosting LLM-Based Vision-and-Language Navigation via Learning Disentangled Reasoning
Bingqian Lin, Yunshuang Nie, Ziming Wei, Jiaqi Chen, Shikui Ma, Jianhua Han, Hang Xu, Xiaojun Chang, Xiaodan Liang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add the Future Imagination task to action prediction (pretrain+finetune) (ablation) | NavCoT LLaMA 2 · R2R Val Unseen subset | SR: 21.11 -> 24.44 | Improved |
| add the Visual Information Filter task to action prediction (pretrain+finetune) (ablation) | NavCoT LLaMA 2 · R2R Val Unseen subset | SPL: 14.82 -> 19.08 | Improved |
| add VIF on top of AP+FI (ablation) | NavCoT LLaMA 2 · R2R Val Unseen subset | SR: 24.44 -> 30.00 | Improved |
| add FI on top of AP+VIF (ablation) | NavCoT LLaMA 2 · R2R Val Unseen subset | SR: 23.33 -> 30.00 | Improved |
| use NavCoT prompt instead of original CoT prompt, zero-shot LLaMA 2 (ablation) | LLaMA 2 · R2R Val Unseen subset | SR: 3.33 -> 16.67 | Improved |
| add in-domain training to NavCoT prompt (ablation) | NavCoT LLaMA 2 · R2R Val Unseen subset | SR: 16.67 -> 53.33 | Improved |
| use LLaMA 2 instead of LLaMA-Adapter backbone (ablation) | NavCoT · R2R Val Unseen subset | SR: 28.89 -> 53.33 | Improved |
| use LLaMA 2 instead of GPT-4 in zero-shot NavCoT (ablation) | NavCoT zero-shot · R2R Val Unseen subset | SR: 15.56 -> 16.67 | Improved |
| let the LLM choose the action instead of CLIP matching the imagination (ablation) | NavCoT LLaMA 2 · R2R Val Unseen subset | SR: 32.23 -> 53.33 | Improved |
| add 12000 R2R augmentation samples | NavCoT LLaMA 2 · R2R Val Unseen | SR: 36.40 -> 40.23 | Improved |
| replace direct action prediction with NavCoT reasoning | LLaMA 2 7B · RxR English Val Unseen | SR: 20.46 -> 24.52 | Improved |
| replace direct action prediction with NavCoT reasoning | LLaMA 2 7B · REVERIE Val Unseen | SR: 3.12 -> 9.20 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- NavGPT — NavGPT: Explicit Reasoning in Vision-and-Language Navigation with Large Language Models (AAAI 2024) · better on 2 of 2 shared results
- DUET — DUET: Cross-Modal Semantic Grounding for Contrastive Zero-Shot Learning (AAAI 2023) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- EvolveNav: Empowering LLM-Based Vision-Language Navigation via Self-Improving Embodied Reasoning (TPAMI 2026 · similarity 0.94)
- NavGPT: Explicit Reasoning in Vision-and-Language Navigation with Large Language Models (AAAI 2024 · similarity 0.93)
- FantasyVLN: Unified Multimodal Chain-of-Thought Reasoning for Vision-and-Language Navigation (CVPR 2026 · similarity 0.91)
- Aux-Think: Exploring Reasoning Strategies for Data-Efficient Vision-Language Navigation (NeurIPS 2025 · similarity 0.91)
- NavBench: Probing Multimodal Large Language Models for Embodied Navigation (NeurIPS 2025 · similarity 0.90)
- Run, Ruminate, and Regulate: A Dual-process Thinking System for Vision-and-Language Navigation (AAAI 2026 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
