arXiv cs.AI· Shaoyi Zheng, Zhongtao Jiang, Yue Peng, Wenbo Lu, Shunchi Zhang, Xi Wang, Shengji Wan·· 12 小时前AI 评分38
DirectWAM：将级联世界-动作模型蒸馏为单步机器人控制预测器
De-Diffusing WAMs: Action-Sufficient Direct Prediction for Robot Control
AI 导读
DirectWAM 通过动作充分蒸馏，把预训练视频与动作专家转为单步预测器，无需随机采样即可保留显式的未来到动作级联。它在四个 LIBERO 套件上平均成功率达 97.1%，RoboTwin 2.0 上达 85.4%，并在三个真实任务中验证有效。相比完整迭代教师模型，其推理延迟降低 28.0 倍。
正文 · 原文
Abstract:Cascade world-action models (WAMs) use predicted visual futures to condition robot actions, but inherit costly iterative inference from diffusion-pretrained experts. We ask what a single-step predicted future must preserve to support accurate single-step action prediction. In the evaluated WAM, resampling video noise has little effect on downstream actions, yet naively truncating both streams to one step substantially reduces task success. Video-space supervision only partially closes this gap, and a local sensitivity analysis shows that vision prediction errors have highly unequal consequences for action accuracy. These findings motivate DirectWAM, which converts pretrained video and action experts into single-step predictors without stochastic sampling through action-sufficient distillation, while retaining the explicit future-to-action cascade. After learning a one-step action model, we train the future predictor by decoding actions from its own predicted futures and backpropagating demonstration action error through the expert, with an optional future-latent reconstruction loss. The resulting objective evaluates the student-generated future through its downstream control consequences rather than relying on representation matching alone. DirectWAM achieves 97.1% average success across four LIBERO suites and 85.4% on RoboTwin 2.0, outperforming Flash-WAM at the same one-evaluation-per-stream budget. This advantage has also extended to three real-world tasks that we designed. Compared with the full iterative teacher, it reduces inference latency by $28.0\times$.
| Subjects: | Robotics (cs.RO); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.28122 [cs.RO] |
| (or arXiv:2610.28122v1 [cs.RO] for this version) | |
| https://doi.org/10.48550/arXiv.2610.28122 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Shaoyi Zheng [view email]
[v1]
Tue, 20 Oct 2026 12:30:58 UTC (15,776 KB)
来源：arXiv cs.AI · arxiv.org