arXiv：cs.LG· Susie Lu, Haonan Chen, Weirui Ye, Yilun Du·· 5 小时前AI 评分47
DriftWorld：通过 Drifting 实现快速世界建模
DriftWorld: Fast World Modeling through Drifting
AI 导读
DriftWorld 是一种基于 drifting 生成模型的动作条件世界模型，训练时学习条件 drift，推理时单次前向传播即可为给定动作序列生成未来观测。在 Bridge-V2、RT-1、Language Table、Push-T 和 Robomimic 上，它以超过 40 fps 运行，比基于扩散模型的基线快 12 倍以上，视觉生成质量相当或更优。
正文
Abstract:Predictive world models enable robots to simulate the visual outcomes of their actions, but state-of-the-art diffusion-based models remain costly because generating each rollout requires multi-step iterative denoising. We introduce DriftWorld, an action-conditioned world model based on drifting generative models. DriftWorld learns a conditional drift during training, enabling it to generate future observations for a given action sequence in a single forward pass during inference. Across Bridge-V2, RT-1, Language Table, Push-T, and Robomimic, DriftWorld runs at over 40 fps and is 12+ times faster than diffusion-based baselines, while matching or improving their visual generation quality. This makes DriftWorld an efficient world model for robot simulation and further enables downstream applications including inference-time action search and offline policy evaluation.
| Comments: | Website at this https URL |
| Subjects: | Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG) |
| Cite as: | arXiv:2607.15065 [cs.RO] |
| (or arXiv:2607.15065v3 [cs.RO] for this version) | |
| https://doi.org/10.48550/arXiv.2607.15065 arXiv-issued DOI via DataCite |
Submission history
From: Susie Lu [view email]
[v1]
Thu, 16 Jul 2026 14:37:43 UTC (11,176 KB)
[v2]
Mon, 3 Aug 2026 21:25:00 UTC (11,176 KB)
[v3]
Fri, 2 Oct 2026 17:56:20 UTC (2,942 KB)
来源：arXiv：cs.LG · arxiv.org