Research map / Vision-language and multimodal / Robot action and navigation
Embodied Navigation Foundation Model
Jiazhao Zhang, Anqi Li, Yunpeng Qi, Minghan Li, Jiahang Liu, Shaoan Wang, Haoran Liu, Gengze Zhou, Yuze Wu, Xingxing LI, Yuxin Fan, Wenjun Li and 5 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 |
|---|---|---|---|
| use temporal-viewpoint indicator tokens instead of viewpoint-history positional embedding (ablation) | NavFoM · VLN-CE RxR Val-Unseen (four views) | SR: 52.3 -> 64.4 | Improved |
| use TVI tokens instead of individually learned special tokens per viewpoint (ablation) | NavFoM · VLN-CE RxR Val-Unseen (four views) | SR: 59.1 -> 64.4 | Improved |
| add learned angle/time MLP projections to indicator tokens (vs handcrafted tokens) (ablation) | NavFoM TVI tokens · VLN-CE RxR Val-Unseen (four views) | SR: 53.6 -> 64.4 | Improved |
| use BATS instead of uniform sampling (budget 2048) (ablation) | NavFoM · VLN-CE RxR Val-Unseen (four views) | SR: 62.4 -> 64.4 | Improved |
| use BATS instead of uniform sampling (budget 1024) (ablation) | NavFoM · VLN-CE RxR Val-Unseen (four views) | SR: 59.7 -> 62.5 | Improved |
| use BATS instead of token merging (budget 2048) (ablation) | NavFoM · VLN-CE RxR Val-Unseen (four views) | SR: 63.2 -> 64.4 | Improved |
| use BATS instead of linear probability sampling (budget 2048) (ablation) | NavFoM · VLN-CE RxR Val-Unseen (four views) | nDTW: 64.8 -> 65.8 | Improved |
| increase cameras from one to four (ablation) | NavFoM · VLN-CE RxR Val-Unseen | SR: 57.4 -> 64.4 | Improved |
| increase cameras from four to six at fixed token budget 2048 (ablation) | NavFoM · VLN-CE RxR Val-Unseen | SR | Hurt |
| co-tune with data from other navigation tasks (50% to 100%) instead of single-task data (ablation) | NavFoM · HM3D-OVON four-view zero-shot (Searching) | SR: 10.3 -> 45.2 | Improved |
| co-tune with data from other navigation tasks instead of single-task data (ablation) | NavFoM · EVT-Bench four-view (Tracking) | SR: 12.6 -> 62.0 | Improved |
| switch from single view to four views (zero-shot for tracking) | NavFoM · EVT-Bench Single Target | SR: 85.0 -> 88.4 | Improved |
Closest papers by meaning
- NavBench: Probing Multimodal Large Language Models for Embodied Navigation (NeurIPS 2025 · similarity 0.90)
- NaVLA$^2$: A Vision-Language-Audio-Action Model for Multimodal Instruction Navigation (AAAI 2026 · similarity 0.89)
- NavGPT: Explicit Reasoning in Vision-and-Language Navigation with Large Language Models (AAAI 2024 · similarity 0.88)
- Image2Sim: Scaling Embodied Navigation via Generative Neural Simulator (NeurIPS 2026 · similarity 0.88)
- Towards Learning a Generalist Model for Embodied Navigation (CVPR 2024 · similarity 0.87)
- C-NAV: Towards Self-Evolving Continual Object Navigation in Open World (NeurIPS 2025 · similarity 0.87)
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.
