Research map / Video / Video generation
Learning Transferable Dynamics Priors from Action to World Modeling
Ze Huang, Zhang Jiahui, Hairuo Liu, Chenxi Zhang, Ran Cheng, Li Zhang
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 |
|---|---|---|---|
| action-conditioned pretraining instead of text-conditioned pretraining (rollout) (ablation) | A2World-sim · LIBERO rollout | PSNR: 26.18 -> 26.64 | Improved |
| action-conditioned pretraining instead of text-conditioned pretraining (rollout) (ablation) | A2World-sim · Real robot rollout (Flexiv Rizon 4S) | PSNR: 24.64 -> 25.95 | Improved |
| add pose-guided history sampling versus no history (ablation) | A2World-sim · LIBERO rollout | PSNR: 25.41 -> 26.64 | Improved |
| add pose-guided history sampling versus sliding window (ablation) | A2World-sim · LIBERO rollout | PSNR: 25.63 -> 26.64 | Improved |
| add pose-guided history sampling versus sliding window (ablation) | A2World-sim · LIBERO rollout | EPE: .3900 -> .3498 | Improved |
| action-to-video pretraining instead of Cosmos initialization (OOD policy) (ablation) | A2World-policy · LIBERO-Plus Spatial | success rate: 80.2 -> 88.5 | Improved |
| action-to-video pretraining instead of text-conditioned pretraining (OOD policy) (ablation) | A2World-policy · LIBERO-Plus Spatial | success rate: 85.8 -> 88.5 | Improved |
| policy-targeted pretraining instead of action-to-video pretraining (OOD) (ablation) | A2World-policy · LIBERO-Plus Spatial | success rate: 88.5 -> 88.6 | No clear effect |
| action-to-video pretraining instead of Cosmos initialization (in-domain) (ablation) | A2World-policy · LIBERO | success rate: 97.0 -> 98.6 | Improved |
| action-to-video pretraining instead of text-conditioned pretraining (in-domain) (ablation) | A2World-policy · LIBERO | success rate: 97.4 -> 98.6 | Improved |
| policy-targeted pretraining instead of action-to-video pretraining (in-domain) (ablation) | A2World-policy · LIBERO | success rate: 98.6 -> 98.8 | No clear effect |
| train video branch jointly instead of freezing it (ablation) | A2World-policy · LIBERO | success rate: 86.2 -> 98.6 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- Cosmos Policy — Cosmos Policy: Fine-Tuning Video Models for Visuomotor Control and Planning (ICLR 2026) · better on 2 of 2 shared results
- SAMPO — SAMPO: Scale-wise Autoregression with Motion Prompt for Generative World Models (NeurIPS 2025) · better on 1 of 1 shared result
- iVideoGPT — iVideoGPT: Interactive VideoGPTs are Scalable World Models (NeurIPS 2024) · better on 1 of 1 shared result
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
- EA-WM: Event-Aware Generative World Model with Structured Kinematic-to-Visual Action Fields (NeurIPS 2026 · similarity 0.91)
- Predictive Inverse Dynamics Models are Scalable Learners for Robotic Manipulation (ICLR 2025 · similarity 0.91)
- VidMan: Exploiting Implicit Dynamics from Video Diffusion Model for Effective Robot Manipulation (NeurIPS 2024 · similarity 0.90)
- Learning an Actionable Discrete Diffusion Policy via Large-Scale Actionless Video Pre-Training (NeurIPS 2024 · similarity 0.90)
- World Action Verifier: Self-Improving World Models via Forward-Inverse Asymmetry (NeurIPS 2026 · similarity 0.89)
- SyncWorld: Visual Calibration Enables World Models as Zero-Shot Simulators (NeurIPS 2026 · similarity 0.89)
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.
