Research map / Vision-language and multimodal / Robot action and navigation
Towards Unified World Models for Visual Navigation via Memory-Augmented Planning and Foresight
Yifei Dong, Fengyi Wu, Guangyu Chen, Lingdong Kong, Xu Zhu, Qiyu Hu, Yuxuan Zhou, Jingdong Sun, Jun-Yan He, Qi Dai, Alexander Hauptmann, Zhi-Qi Cheng
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 |
|---|---|---|---|
| add intra-step memory (ablation) | UniWM · Go Stanford | Success Rate: 0.71 -> 0.73 | Improved |
| add cross-step memory on top of intra-step memory (ablation) | UniWM · ReCon | Success Rate: 0.85 -> 0.93 | Improved |
| add both memory banks (ablation) | UniWM · HuRoN | Success Rate: 0.70 -> 0.76 | Improved |
| train planner and world model jointly instead of separately (ablation) | UniWM · Navigation avg | Success Rate: 0.65 -> 0.71 | Improved |
| train planner and world model jointly instead of separately (ablation) | UniWM · Visualization avg | SSIM: 0.443 -> 0.457 | Improved |
| interleave planner and world-model substeps instead of predicting both (ablation) | UniWM · Go Stanford | Success Rate: 0.65 -> 0.71 | Improved |
| use 1 x 784 tokens instead of 2 x 484 tokens (ablation) | UniWM · Navigation avg (no memory) | Success Rate: 0.55 -> 0.71 | Improved |
| use 2 x 625 tokens instead of 1 x 784 tokens (ablation) | UniWM · Navigation avg (no memory) | Success Rate: 0.71 -> 0.68 | Hurt |
| use 5 memory layers instead of 32 (ablation) | UniWM · Go Stanford | Success Rate: 0.58 -> 0.75 | Improved |
| use 5 memory layers instead of 3 (ablation) | UniWM · Go Stanford | ATE: 0.27 -> 0.22 | Improved |
| use 7 memory layers instead of 5 (ablation) | UniWM · SCAND | Success Rate: 0.68 -> 0.69 | Mixed |
| add goal conditioning in visualization substep (ablation) | UniWM · TartanDrive (zero-shot) | Success Rate: 0.33 -> 0.35 | Improved |
Closest papers by meaning
- Planning from Imagination: Episodic Simulation and Episodic Memory for Vision-and-Language Navigation (AAAI 2025 · similarity 0.90)
- ImagineNav: Prompting Vision-Language Models as Embodied Navigator through Scene Imagination (ICLR 2025 · similarity 0.89)
- Navigation World Models (CVPR 2025 · similarity 0.89)
- FantasyVLN: Unified Multimodal Chain-of-Thought Reasoning for Vision-and-Language Navigation (CVPR 2026 · similarity 0.88)
- NavGPT: Explicit Reasoning in Vision-and-Language Navigation with Large Language Models (AAAI 2024 · similarity 0.88)
- Drift-Resistant Navigation World Model with Anchored Epipolar Guidance (NeurIPS 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.
