arXiv：cs.LG（机器学习，全量分类）· Seungyeon Kim, Junhoo Lee, Baekseung Kim, Minkyu Kim, Nojun Kwak·· 21 小时前AI 评分39
面向世界动作模型的完成感知引导（CAG）
Completion Aware Guidance for World Action Models
AI 导读
研究者提出 Completion Aware Guidance（CAG），一种免训练采样方法，引导世界动作模型（WAM）生成任务完成所需的过渡，缓解短块控制下反复偏向局部合理延续的问题。在代表性 WAM 上，CAG 将 RoboTwin 2.0 子集成功率从 64% 提升至 70%，零样本仿真从 69% 提升至 75%，并将任务未完成想象从 79% 降至 40%。
正文
Abstract:World Action Models (WAMs) predict visual futures and robot actions, yet they remain susceptible to task-incomplete imagination, where plausible, action-consistent predictions omit the transition needed for task completion. In this paper, we show that this failure is not inherent to the world model backbone, but emerges when adapted for short-chunk control, which can repeatedly favor plausible local continuations over task-completing transitions. To address this, we introduce Completion Aware Guidance (CAG), a training-free sampling method that guides generation toward task completion. Across representative WAMs, CAG improves success from 64% to 70% on a RoboTwin 2.0 subset and from 69% to 75% in zero-shot simulation, while reducing task-incomplete imagination from 79% to 40%.
| Subjects: | Robotics (cs.RO); Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.01559 [cs.RO] |
| (or arXiv:2610.01559v1 [cs.RO] for this version) | |
| https://doi.org/10.48550/arXiv.2610.01559 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Seungyeon Kim [view email]
[v1]
Thu, 1 Oct 2026 12:27:29 UTC (3,337 KB)
来源：arXiv：cs.LG（机器学习，全量分类） · arxiv.org