Research map / Vision-language and multimodal / Robot action and navigation
Rethinking the Embodied Gap in Vision-and-Language Navigation: A Holistic Study of Physical and Visual Disparities
Liuyi Wang, Xinyuan Xia, Hui Zhao, Hanqing Wang, Tai Wang, Yilun Chen, Chengju Liu, Qijun Chen, Jiangmiao Pang
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 |
|---|---|---|---|
| transfer VLN-CE models zero-shot to physical VLN-PE | Seq2Seq-Full, CMA-Full, NaVid · R2R (H1 humanoid) | SR: declines of 10%, 16%, and 18% | Hurt |
| raise camera height to 1.8m to match H1 (ablation) | CMA · R2R val-unseen (H1 humanoid) | SR: 10.31 -> 13.58 | Improved |
| fine-tune VLN-CE weights on VLN-PE data (CMA+) | CMA-Full · R2R val-unseen (H1 humanoid) | SR: 16.04 -> 23.31 | Improved |
| use diffusion policy head (RDP) instead of CMA when training from scratch | CMA · R2R val-unseen (H1 humanoid) | SR: 18.78 -> 21.98 | Improved |
| fine-tune on in-domain GRU-VLN10 scenes (ablation) | RDP · GRU-VLN10 val-unseen (H1 humanoid) | SR: 26.19 -> 28.52 | Improved |
| fine-tune on in-domain GRU-VLN10 scenes (ablation) | CMA-CLIP · GRU-VLN10 val-unseen (H1 humanoid) | SR: 15.31 -> 22.46 | Improved |
| use data collected with locomotion controller when evaluating with controller (ablation) | CMA · R2R val-unseen, CMA from scratch (controller engagement) | SR: 13.51 -> 18.78 | Improved |
| evaluate with locomotion controller without training on controller data (ablation) | CMA · R2R val-unseen, CMA from scratch (controller engagement) | SR: 21.31 -> 13.51 | Hurt |
| train with humanoid, quadruped and wheeled data instead of humanoid data only (ablation) | CMA · R2R val-unseen, CMA cross-embodiment (humanoid) | SR: 18.78 -> 19.30 | No clear effect |
| add VLN-CE data to all three robots' training data (ablation) | CMA · R2R val-unseen, CMA cross-embodiment (humanoid) | SR: 19.30 -> 26.87 | Improved |
| add VLN-CE data to all three robots' training data (ablation) | CMA · R2R val-unseen, CMA cross-embodiment (quadruped) | SR: 17.15 -> 23.83 | Improved |
| reduce light from DL5000 to DL300 | NaVid (RGB only) · R2R val-unseen (H1 humanoid) | SR: 22.42 -> 9.95 | Hurt |
Closest papers by meaning
- CoT-VLNBench: A Benchmark for Visual Chain-of-Thought Reasoning in Vision-Language-Navigation Robots (AAAI 2026 · similarity 0.91)
- Human-Aware Vision-and-Language Navigation: Bridging Simulation to Reality with Dynamic Human Interactions (NeurIPS 2024 · similarity 0.90)
- Towards Realistic UAV Vision-Language Navigation: Platform, Benchmark, and Methodology (ICLR 2025 · similarity 0.89)
- ImagineNav: Prompting Vision-Language Models as Embodied Navigator through Scene Imagination (ICLR 2025 · similarity 0.88)
- NavBench: Probing Multimodal Large Language Models for Embodied Navigation (NeurIPS 2025 · similarity 0.88)
- PhysVLM: Enabling Visual Language Models to Understand Robotic Physical Reachability (CVPR 2025 · 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.
