Research map / Video / Video generation
DVG-WM: Disentangled Video Generation Enables Efficient Embodied World Model for Robotic Manipulation
Ziyu Shan, Zhenyu Wu, Xiaofeng Wang, Zheng Zhu, Ziwei Wang
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 latent degradation instead of pixel degradation (ablation) | DVG-WM refinement · LIBERO | SSIM: 0.768 -> 0.783 | Improved |
| add latent degradation instead of pixel degradation (ablation) | DVG-WM refinement · LIBERO | FVD: 168.54 -> 152.36 | Improved |
| use flow matching from low-res latents instead of naive cascading conditioning (ablation) | DVG-WM · LIBERO | FVD: 187.69 -> 152.36 | Improved |
| use 256x384 preview resolution instead of 360p (ablation) | DVG-WM preview stage · LIBERO | PSNR: 18.812 -> 20.019 | Improved |
| add preview stage to refine-only model (ablation) | DVG-WM · LIBERO | FVD: 169.29 -> 152.36 | Improved |
| add refinement stage to preview-only model (ablation) | DVG-WM · LIBERO | LPIPS: 0.341 -> 0.120 | Improved |
| increase refinement steps K from 1 to 4 (ablation) | DVG-WM · LIBERO | FVD: 245.83 -> 152.36 | Improved |
| increase refinement steps K from 4 to 8 (ablation) | DVG-WM · LIBERO | FVD: 152.36 -> 150.42 | No clear effect |
| use disentangled two-stage DVG-WM instead of CogVideoX-5B | world model · LIBERO | FVD: 171.24 -> 152.36 | Improved |
| use DVG-WM instead of LVP-14B for speed | world model · LIBERO | inference time: 354.2s -> 88.7s | Improved |
| use DVG-WM instead of LVP-14B | world model · LIBERO | object-level accuracy: 80% -> 89% | Improved |
| use DVG-WM instead of CogVideoX with action expert | real-world manipulation · 9 real-world tasks | success rate: 34.0 -> 62.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- CogVideoX-5B — CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer (ICLR 2025) · better on 7 of 7 shared results
- LongScape — LongScape: Advancing Long-Horizon Embodied World Models with Context-Aware MoE (NeurIPS 2026) · better on 3 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- LongScape: Advancing Long-Horizon Embodied World Models with Context-Aware MoE (NeurIPS 2026 · similarity 0.90)
- EnerVerse: Envisioning Embodied Future Space for Robotics Manipulation (NeurIPS 2025 · similarity 0.90)
- DiT4DiT: Jointly Modeling Video Dynamics and Actions for Generalizable Robot Control (NeurIPS 2026 · similarity 0.89)
- RoboScape: Physics-informed Embodied World Model (NeurIPS 2025 · similarity 0.89)
- EA-WM: Event-Aware Generative World Model with Structured Kinematic-to-Visual Action Fields (NeurIPS 2026 · similarity 0.89)
- WorldSimBench: Towards Video Generation Models as World Simulators (ICML 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.
