Research map / 3D vision / 3D generation
MVISTA-4D: View-Consistent 4D World Model with Test-Time Action Inference for Robotic Manipulation
Jiaxu Wang, JIANG Yicheng, Tianlun HE, Jingkai SUN, Qiang Zhang, Jiahang Cao, Zesen Gan, Mingyuan Sun, Qiming Shao, Xiangyu Yue
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 geometry-aware deformable cross-view attention (vs none) (ablation) | MVISTA-4D · RoboTwin 4D generation | AbRel (x10^2): 3.34 -> 2.60 | Improved |
| use deformable cross-view attention instead of simple epipolar attention (ablation) | MVISTA-4D · RoboTwin 4D generation | EMD (x10^2): 12.50 -> 9.90 | Improved |
| add cross-modality fusion between RGB and depth (ablation) | MVISTA-4D · RoboTwin 4D generation | PSNR: 20.16 -> 22.91 | Improved |
| optimize trajectory latent at test time instead of using action-head prior (ablation) | MVISTA-4D · RLBench manipulation | success rate (%): 72.5 -> 72.6 | Improved |
| add residual IDM on top of decoded trajectory prior (ablation) | MVISTA-4D · RLBench manipulation | success rate (%): 69.0 -> 72.6 | Improved |
| use residual IDM instead of full point-based IDM (ablation) | MVISTA-4D · RLBench manipulation | success rate (%): 68.8 -> 72.6 | Improved |
| add cross-view and cross-modality fusion (ablation) | MVISTA-4D · RLBench manipulation | success rate (%): 66.5 -> 72.6 | Improved |
| add TCN-VAE trajectory latent (ablation) | MVISTA-4D · RoboTwin manipulation | success rate (%): 41.5 -> 43.0 | Improved |
| fuse depth via token-level modality fusion instead of channel concatenation (ablation) | MVISTA-4D · RLBench manipulation | success rate (%): 66.1 -> 72.6 | Improved |
| initialize trajectory latent with action-head prediction (+ActionInit) (ablation) | MVISTA-4D · RLBench manipulation | success rate (%): 72.6 -> 76.5 | Improved |
| +ActionInit reduces optimization to 10-15 steps (ablation) | MVISTA-4D · RLBench inference latency | total time (s): 58s -> 47s | Improved |
| use MVISTA-4D instead of UniPi* for appearance | RLBench 4D generation · RLBench | PSNR: 23.88 -> 23.31 | Hurt |
Closest papers by meaning
- Predictive Inverse Dynamics Models are Scalable Learners for Robotic Manipulation (ICLR 2025 · similarity 0.90)
- DynaRend: Learning 3D Dynamics via Masked Future Rendering for Robotic Manipulation (NeurIPS 2025 · similarity 0.89)
- ConsisVLA-4D: Advancing Spatiotemporal Consistency in Efficient 3D-Perception and 4D-Reasoning for Robotic Manipulation (CVPR 2026 · similarity 0.89)
- EnerVerse: Envisioning Embodied Future Space for Robotics Manipulation (NeurIPS 2025 · similarity 0.89)
- World Motion Models: Flexible Sequence Modeling of SE(3) Trajectories (NeurIPS 2026 · similarity 0.88)
- Localizing, Structuring, and Rendering: Bridging 3D and 2D Vision-Language-Action Models for Robotic Manipulation (CVPR 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.
