Research map / Video / Video generation
Astra: General Interactive World Model with Autoregressive Denoising
Yixuan Zhu, Feng Jiaqi, Wenzhao Zheng, Yuan Gao, Xin Tao, Pengfei Wan, Jiwen Lu, Jie Zhou
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 noise-as-mask corruption of history frames during training (ablation) | Astra · Astra-Bench | instruction following: 0.359 -> 0.669 | Improved |
| add noise-as-mask corruption of history frames during training (ablation) | Astra · Astra-Bench | subject consistency: 0.903 -> 0.939 | Improved |
| add action-free guidance (scale 3.0) at inference (ablation) | Astra · Astra-Bench | instruction following: 0.545 -> 0.669 | Improved |
| add action-free guidance (ablation) | Astra · Astra-Bench | subject consistency: 0.841 -> 0.939 | Improved |
| use ACT-Adapter (identity-initialized linear adapter plus additive action features) instead of cross-attention adapter (ablation) | Astra · Astra-Bench | instruction following: 0.642 -> 0.669 | Improved |
| use ACT-Adapter instead of cross-attention adapter (ablation) | Astra · Astra-Bench | motion smoothness: 0.948 -> 0.989 | Improved |
| add Mixture of Action Experts (ablation) | Astra · Astra-Bench | instruction following: 0.651 -> 0.669 | Improved |
| add Mixture of Action Experts (ablation) | Astra · Astra-Bench | imaging quality: 0.727 -> 0.747 | Improved |
| Astra vs MatrixGame | MatrixGame · Astra-Bench | imaging quality: 0.748 -> 0.747 | Hurt |
| Astra vs Yume on camera rotation error | Yume · Astra-Bench | camera rotation error: 2.20 -> 1.23 | Improved |
Closest papers by meaning
- LongScape: Advancing Long-Horizon Embodied World Models with Context-Aware MoE (NeurIPS 2026 · similarity 0.89)
- Prediction with Action: Visual Policy Learning via Joint Denoising Process (NeurIPS 2024 · similarity 0.89)
- SAMPO: Scale-wise Autoregression with Motion Prompt for Generative World Models (NeurIPS 2025 · similarity 0.89)
- DiT4DiT: Jointly Modeling Video Dynamics and Actions for Generalizable Robot Control (NeurIPS 2026 · similarity 0.87)
- PlaySlot: Learning Inverse Latent Dynamics for Controllable Object-Centric Video Prediction and Planning (ICML 2025 · similarity 0.87)
- DiffusionDrive: Truncated Diffusion Model for End-to-End Autonomous Driving (CVPR 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.
