Research map / Vision-language and multimodal / Robot action and navigation
RAE-NWM: Navigation World Model in Dense Visual Representation Space
Mingkun Zhang, wangtian shen, Fan Zhang, Haijian Qin, Zihao Pei, Ziyang Meng
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 |
|---|---|---|---|
| learned time-driven gate instead of simple addition (ablation) | RAE-NWM · SACSoN planning | ATE / RPE: 3.34 -> 2.91 ATE, 0.79 -> 0.70 RPE | Improved |
| learned time-driven gate instead of MLP fusion (ablation) | RAE-NWM · SACSoN planning | ATE / RPE: 3.54 -> 2.91 ATE, 0.83 -> 0.70 RPE | Improved |
| learned gate instead of scheduled gate (ablation) | RAE-NWM · SACSoN planning | ATE / RPE: 3.31 -> 2.91 ATE, 0.78 -> 0.70 RPE | Improved |
| flow matching generative objective instead of deterministic DINO-token regression (ablation) | RAE-NWM · SACSoN planning | ATE / RPE: 3.14 -> 2.91 ATE, 0.75 -> 0.70 RPE | Improved |
| flow matching instead of deterministic DINO-token regression (ablation) | RAE-NWM · RECON planning | RPE: 0.36 vs. 0.37 | No clear effect |
| DINOv2 dense representation space instead of SD-VAE latent (ablation) | RAE-NWM · long-horizon rollout | LPIPS | Mixed |
| add DDT head (ablation) | RAE-NWM · long-horizon rollout | LPIPS | Improved |
| RAE-NWM instead of NWM on direct 16s prediction | RAE-NWM vs NWM · SACSoN direct prediction 16s | FID: 33.06 -> 15.90 | Improved |
| RAE-NWM instead of NWM on direct 16s prediction | RAE-NWM vs NWM · SACSoN direct prediction 16s | LPIPS: 0.470 -> 0.349 | Improved |
| RAE-NWM instead of NWM in planning on RECON | RAE-NWM vs NWM · RECON planning (2s) | ATE: 1.13 -> 1.36 | Hurt |
| RAE-NWM instead of NWM in planning on SACSoN | RAE-NWM vs NWM · SACSoN planning (2s) | ATE: 4.12 -> 2.91 | Improved |
| RAE-NWM in Habitat versus One-Step WM | RAE-NWM vs One-Step WM · Habitat image-goal navigation | SR / SPL: SR 72.67 -> 78.95, SPL 69.10 -> 63.58 | Mixed |
Closest papers by meaning
- Navigation World Models (CVPR 2025 · similarity 0.91)
- Image2Sim: Scaling Embodied Navigation via Generative Neural Simulator (NeurIPS 2026 · similarity 0.88)
- Drift-Resistant Navigation World Model with Anchored Epipolar Guidance (NeurIPS 2026 · similarity 0.87)
- StateSpaceDiffuser: Bringing Long Context to Diffusion World Models (NeurIPS 2025 · similarity 0.87)
- WorldVLN: Autoregressive World Action Model for Aerial Vision-Language Navigation (NeurIPS 2026 · 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.
