Research map / Vision-language and multimodal / Multimodal LLMs
EvolveNav: Empowering LLM-Based Vision-Language Navigation via Self-Improving Embodied Reasoning
Bingqian Lin, Yunshuang Nie, Khun Loun Zai, Ziming Wei, Mingfei Han, Rongtao Xu, Minzhe Niu, Jianhua Han, Hanwang Zhang, Liang Lin, Bokui Chen, Cewu Lu and 1 more
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 formalized CoT SFT (Stage 1) (ablation) | NaviLLM* baseline · R2R val-unseen (IL) | SR: 46.90 -> 49.62 | Improved |
| add self-enriched CoT labels in Stage 2 (ablation) | CoT SFT model · R2R val-unseen (IL) | SR: 49.62 -> 50.47 | Improved |
| add self-reflective auxiliary task in Stage 2 (ablation) | CoT SFT model · R2R val-unseen (IL) | SR: 49.62 -> 50.51 | Improved |
| combine both Stage 2 components (ablation) | CoT SFT model · R2R val-unseen (IL) | SR: 49.62 -> 51.15 | Improved |
| add self-enriched CoT labels (ablation) | CoT SFT model · R2R val-unseen (IL) | OSR: 59.44 -> 57.48 | Hurt |
| use formalized direction-landmark CoT instead of free-form CoT (ablation) | EvolveNav · R2R val-unseen | SR: 69.39 -> 71.26 | Improved |
| use only direction CoT instead of direction and landmark (ablation) | EvolveNav · R2R val-unseen | SR: 71.26 -> 67.05 | Hurt |
| use only landmark CoT instead of direction and landmark (ablation) | EvolveNav · R2R val-unseen | SR: 71.26 -> 68.32 | Hurt |
| use diverse extracted landmarks instead of those mentioned in instruction (ablation) | EvolveNav · R2R val-unseen | SR: 69.77 -> 71.26 | Improved |
| use formalized CoT instead of free-form CoT for inference speed (ablation) | EvolveNav · R2R | CoT generation time per step: ~7.8s -> ~2.5s | Improved |
| apply EvolveNav on top of NaviLLM* | NaviLLM* · CVDN val-unseen | Goal Progress: 5.53 -> 6.21 | Improved |
| apply EvolveNav on top of NaviLLM* (cross-task) | NaviLLM* · R2R cross-task | SR: 68.37 -> 68.07 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- VLN-PETL — VLN-PETL: Parameter-Efficient Transfer Learning for Vision-and-Language Navigation (ICCV 2023) · better on 1 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
- FantasyVLN: Unified Multimodal Chain-of-Thought Reasoning for Vision-and-Language Navigation (CVPR 2026 · similarity 0.93)
- Aux-Think: Exploring Reasoning Strategies for Data-Efficient Vision-Language Navigation (NeurIPS 2025 · similarity 0.92)
- NavGPT: Explicit Reasoning in Vision-and-Language Navigation with Large Language Models (AAAI 2024 · similarity 0.91)
- Run, Ruminate, and Regulate: A Dual-process Thinking System for Vision-and-Language Navigation (AAAI 2026 · similarity 0.90)
- Insight-V: Exploring Long-Chain Visual Reasoning with Multimodal Large Language Models (CVPR 2025 · similarity 0.90)
- Agent0-VL: Exploring Self-Evolving Agent for Tool-Integrated Vision-Language Reasoning (ICML 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.
