Research map / Vision-language and multimodal / Multimodal LLMs
CoNav: Collaborative Cross-Modal Reasoning for Embodied Navigation
Haihong Hao, Mingfei Han, Changlin Li, Zhihui Li, Xiaojun Chang
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 Cross-Modal Belief Alignment fine-tuning on top of sharing 3D-text hypotheses (ablation) | CoNav · R2R Val-U | SPL: 56 -> 62 | Improved |
| add Cross-Modal Belief Alignment fine-tuning (ablation) | CoNav · ScanQA Val-U | ROUGE: 35.7 -> 38.9 | Improved |
| share 3D-text hypotheses without alignment to image-text agent (ablation) | image-text agent · R2R Val-U | SPL: 57 -> 56 | Hurt |
| share 3D-text hypotheses without alignment (ablation) | image-text agent · ScanQA Val-U | ROUGE: 36.7 -> 35.7 | Hurt |
| add 3D-text model guidance with alignment to image-text agent (ablation) | image-text agent · R2R Val-U | SR: 64 -> 69 | Improved |
| replace post-process CEO alignment (Qwen2.5-7B) by trained alignment (ablation) | CoNav · R2R Val-U | SPL: 51 -> 62 | Improved |
| train alignment instead of post-process CEO alignment (ablation) | CoNav · ScanQA Val-U | ROUGE: 35.2 -> 38.9 | Improved |
| use model-level fusion instead of feature-level fusion (ablation) | CoNav · ScanQA Val-U | ROUGE: 31.7 -> 38.9 | Improved |
| use model-level fusion instead of feature-level fusion (ablation) | CoNav · ScanQA Val-U | METEOR: 13.9 -> 15.2 | Improved |
| add curriculum stage 2 (3D-FRONT) after stage 1 for 3D-text model (ablation) | 3D-text model · ScanQA Val-U | METEOR: 8.39 -> 10.1 | Improved |
| add curriculum stage 3 (real scans) after stages 1-2 (ablation) | 3D-text model · ScanQA Val-U | METEOR: 10.1 -> 14.5 | Improved |
| use point-cloud text model alone vs image-text model alone for spatial reasoning (ablation) | single-modality models · ScanQA Val-U | METEOR: 14.3 -> 14.5 | Improved |
Closest papers by meaning
- Image2Sim: Scaling Embodied Navigation via Generative Neural Simulator (NeurIPS 2026 · similarity 0.89)
- NavGPT: Explicit Reasoning in Vision-and-Language Navigation with Large Language Models (AAAI 2024 · similarity 0.88)
- From Correspondence to Actions: Human-Like Multi-Image Spatial Reasoning in Multi-modal Large Language Models (ICML 2026 · similarity 0.88)
- BeliefMapNav: 3D Voxel-Based Belief Map for Zero-Shot Object Navigation (NeurIPS 2025 · similarity 0.88)
- Seeing from Another Perspective: Evaluating Multi-View Understanding in MLLMs (AAAI 2026 · similarity 0.88)
- FantasyVLN: Unified Multimodal Chain-of-Thought Reasoning for Vision-and-Language Navigation (CVPR 2026 · similarity 0.88)
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.
