arXiv cs.AI· Zhexi Luo, Zhixuan Xu, Rui Yang, Zeyu Xiao, Yiwen Hou, Lin Shao·· 2 天前AI 评分31
DexMimic：通过价值引导重置分配实现通用灵巧手-物体跟踪策略
DexMimic: A Generalist Dexterous Hand-Object Tracking Policy via Value-Guided Reset Allocation
AI 导读
DexMimic 是一种跨多样人手-物体交互轨迹联合训练的通用灵巧手-物体跟踪策略，通过 Value-Guided Reset Allocation 在线调整初始化分布，优先训练困难阶段并选择高期望跟踪回报的重置状态。实验显示其跟踪成功率与训练效率提升、微调更快，并具备对未见轨迹和物体的零样本泛化潜力，训练集越大效率与泛化越好，已在真实灵巧操作平台上验证。
正文 · 原文
Abstract:Achieving human-like dexterous manipulation remains a central challenge in robotics. Learning from human hand-object interaction (HOI) datasets offers a promising pathway to diverse dexterous skills, yet existing physics-based approaches train specialist policies for individual trajectories, thereby limiting skill reuse at scale. We introduce DexMimic, a generalist dexterous hand-object tracking policy jointly trained across diverse HOI trajectories. Efficient joint training requires allocating each rollout to both an undertrained phase and a starting state that supports continued tracking. DexMimic realizes this insight through Value-Guided Reset Allocation, which adapts the initialization distribution online to prioritize difficult phases and select reset states with higher expected tracking returns. We further introduce a benchmark spanning training, fine-tuning, and generalization. Experiments show improved tracking success and training efficiency, faster fine-tuning, and the potential for zero-shot generalization to unseen trajectories and objects; larger training sets further improve efficiency and generalization. Finally, we validate DexMimic on a real-world dexterous manipulation platform.
| Comments: | 13 pages, including 4 pages of supplementary material. v2: Updated author affiliations; scientific content unchanged. Project page: this https URL |
| Subjects: | Robotics (cs.RO); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.19089 [cs.RO] |
| (or arXiv:2610.19089v2 [cs.RO] for this version) | |
| https://doi.org/10.48550/arXiv.2610.19089 arXiv-issued DOI via DataCite |
Submission history
From: Zhexi Luo [view email]
[v1]
Tue, 13 Oct 2026 19:29:33 UTC (12,971 KB)
[v2]
Thu, 15 Oct 2026 06:00:59 UTC (12,971 KB)
来源：arXiv cs.AI · arxiv.org