[论文] DexCompose: Reusing Dexterous Policies for Multi-Task Manipulation wit...

论文概要

研究领域: CV 作者: Dihong Huang, Zhenyu Wei, Zhuxiu Xu 发布时间: 2026-06-26 arXiv: 2606.28323

中文摘要

灵巧操作策略可以解决单个技能，但将它们组合起来用单手执行多任务仍然具有挑战性。在现有操作技能之上添加新任务时，往往会对重叠的手指和接触模式产生冲突需求，导致在保持已有操作结果与执行新操作之间产生破坏性干扰。我们提出 DexCompose，一种角色感知的残差组合框架，通过显式的手指级动作所有权，重用预训练的灵巧策略进行多任务操作。给定两个预训练的全手策略，DexCompose 首先收集第一个技能的成功后任务状态，并对候选手指掩码执行释放测试，识别哪些手指对于维持已建立的技能状态是必要的。然后训练两个非对称残差模块：用于任务保持的有界残差稳定器，以及仅在分配给新任务的动作子空间内自适应冻结下游策略的上下文感知残差。我们在16个复合灵巧操作任务上评估该框架，涵盖四个物体保持技能和四个下游交互。DexCompose 达到77.4%的平均复合成功率，表明结构化的动作所有权与双残差设计为超越传统策略链的灵巧技能组合提供了有前景的方向。

原文摘要

Dexterous manipulation policies can solve individual skills, but composing them to perform multiple tasks with a single hand remains challenging. Adding a new task on top of an existing manipulation skill often imposes conflicting demands on overlapping fingers and contact modes, causing destructive interference between preserving an existing manipulation outcome and executing a new one. We propose DexCompose, a role-aware residual composition framework that reuses pretrained dexterous policies for multi-task manipulation through explicit finger-level action ownership. Given two pretrained full-hand policies, DexCompose first collects successful post-task states from the first skill and performs release tests over candidate finger masks to identify which fingers are necessary for maintaini...

--- *自动采集于 2026-06-30*

#论文 #arXiv #CV #小凯