Research map / Vision-language and multimodal / Multimodal LLMs
MM-Nav: Multi-View VLA Model for Robust Visual Navigation via Multi-Expert Learning
Tianyu Xu, Jiawei Chen, Jiazhao Zhang, Wenyao Zhang, Zekun Qi, Minghan Li, Jiahang Liu, Lu Yue, Zhizheng Zhang, HE WANG
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 |
|---|---|---|---|
| co-train with large-scale real-world VQA data (ablation) | MM-Nav · InternVLA-N1 S1 benchmark | SR: Commercial 73.9 -> 86.7; Home 76.4 -> 79.9 (text: >20% drop without VQA) | Improved |
| co-train with VQA data in real-world trials (ablation) | MM-Nav · real-world thin/irregular obstacles | SR | Improved |
| increase VQA ratio from 0% to 100% (ablation) | MM-Nav latent space · sim vs real hidden-state alignment | gap-axis distance: 1.86 -> 0.38 | Improved |
| use 360 multi-view input instead of front view only (ablation) | MM-Nav · InternVLA-N1 S1 benchmark | SR: Home 79.9 -> 86.3; Commercial 86.7 -> 89.9 | Improved |
| train VLA on mixed data from all three RL experts instead of one expert (ablation) | MM-Nav VLA · IsaacLab capability scenes | SR: Squeezing 64 -> 71; Avoiding 62 -> 68; Reaching 70 -> 80 | Improved |
| VLA student distilled from RL experts vs the experts | Mixed-VLA · IsaacLab capability scenes | SR: Reaching 65 -> 80, Avoiding 58 -> 68, Squeezing 73 -> 71 | Mixed |
| single mixed-scene RL expert instead of specialized experts (ablation) | RL expert · IsaacLab capability scenes | SR: Squeezing 73 -> 58; Reaching 65 -> 45; Avoiding 58 -> 42 | Hurt |
| capability-balanced data aggregation instead of unbalanced (ablation) | MM-Nav iterative training · IsaacLab capability scenes | SR | Mixed |
| MM-Nav vs NavDP on IsaacLab | MM-Nav · IsaacLab squeezing scene | SR: 18 -> 71 | Improved |
| MM-Nav vs strongest baseline SIDP on InternVLA-N1 S1 | MM-Nav · InternVLA-N1 S1 benchmark | SR: Home 63.2 -> 86.3; Commercial 81.2 -> 89.9 | Improved |
Closest papers by meaning
- Towards Long-Horizon Vision-Language Navigation: Platform, Benchmark and Method (CVPR 2025 · similarity 0.88)
- Dynam3D: Dynamic Layered 3D Tokens Empower VLM for Vision-and-Language Navigation (NeurIPS 2025 · similarity 0.87)
- NavGPT: Explicit Reasoning in Vision-and-Language Navigation with Large Language Models (AAAI 2024 · similarity 0.87)
- VLN-Video: Utilizing Driving Videos for Outdoor Vision-and-Language Navigation (AAAI 2024 · similarity 0.87)
- MMT-Bench: A Comprehensive Multimodal Benchmark for Evaluating Large Vision-Language Models Towards Multitask AGI (ICML 2024 · similarity 0.87)
- OVER-NAV: Elevating Iterative Vision-and-Language Navigation with Open-Vocabulary Detection and StructurEd Representation (CVPR 2024 · similarity 0.86)
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.
