Research map / Vision-language and multimodal / Multimodal LLMs
CoT-VLNBench: A Benchmark for Visual Chain-of-Thought Reasoning in Vision-Language-Navigation Robots
Xiao Zhao, Chang Liu, Ruiteng Ji, Zheyuan Zhang, Mingxu Zhu, Linna Song, Zhe Ren, Luo Qingliang, YuHang Gao, Zhaolong Du, Chufan Guo, Kuifeng Su
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces CoT-VLNBench, a benchmark for visual chain-of-thought reasoning in vision-language-navigation robots, motivated by the lack of vision-language task support in robot-centric datasets.
Problem
Existing robot-centric datasets focus on traditional 3D tasks such as perception and prediction and lack adequate support for vision-language navigation (VLN) reasoning in embodied settings.
Approach
A benchmark targeting visual chain-of-thought reasoning for vision-language-navigation robots, aimed at interpretable navigation in complex environments. Full method details are unavailable from the abstract.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add CoT reasoning to vision-language training (ablation) | Qwen2.5-VL 7B · CoT-VLNBench val | avg L2 error (m): 0.33 -> 0.29 | Improved |
| Add CoT reasoning to vision-language training (ablation) | All four VLMs on average · CoT-VLNBench val | avg L2 error (m): 0.37 -> 0.34 | Improved |
| Add visual input to language-only training (ablation) | All four VLMs on average · CoT-VLNBench val | avg L2 error (m): 0.42 -> 0.37 | Improved |
| Add visual input and CoT to language-only model (ablation) | Qwen2.5-VL 7B · CoT-VLNBench val | L2 error at 3s (m): 0.62 -> 0.47 | Improved |
| Add CoT reasoning (ablation) | InternVL3-8B · CoT-VLNBench val | avg L2 error (m): 0.34 -> 0.31 | Improved |
| Add CoT reasoning (ablation) | Janus-Pro 7B · CoT-VLNBench val | avg L2 error (m): 0.42 -> 0.39 | Improved |
| Add CoT reasoning (ablation) | LLaVA-1.6 7B · CoT-VLNBench val | avg L2 error (m): 0.40 -> 0.35 | Improved |
| Use CoT-VLN (Qwen2.5-VL) instead of SparseDrive | Non-VLM trajectory predictors · CoT-VLNBench val | avg L2 error (m): 0.42 -> 0.29 | Improved |
| Use CoT-VLN (Qwen2.5-VL) instead of DiffusionPlanner | Non-VLM trajectory predictors · CoT-VLNBench val | avg L2 error (m): 0.37 -> 0.29 | Improved |
| Use Qwen2.5-VL backbone instead of LLaVA-1.6 | CoT-VLN · CoT-VLNBench val | GPT-4.1 CoT score vs L2: score 66.48 -> 75.67; L2 0.35 -> 0.29 | Mixed |
Experimental setup
Benchmarks: CoT-VLNBench
Closest papers by meaning
- FantasyVLN: Unified Multimodal Chain-of-Thought Reasoning for Vision-and-Language Navigation (CVPR 2026 · similarity 0.94)
- Aux-Think: Exploring Reasoning Strategies for Data-Efficient Vision-Language Navigation (NeurIPS 2025 · similarity 0.92)
- CoT-VLA: Visual Chain-of-Thought Reasoning for Vision-Language-Action Models (CVPR 2025 · similarity 0.91)
- Seeing Across Views: Benchmarking Spatial Reasoning of Vision-Language Models in Robotic Scenes (ICLR 2026 · similarity 0.91)
- NavGPT: Explicit Reasoning in Vision-and-Language Navigation with Large Language Models (AAAI 2024 · similarity 0.91)
- NavBench: Probing Multimodal Large Language Models for Embodied Navigation (NeurIPS 2025 · 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.
