# DONIC双臂灵巧跟踪策略发布，零样本泛化成功率达86.79%

> 原文链接：https://robotics.hellobike.com/papers/2026-10-21-2610-29997-49eba9cd
> 频道：papers | 更新日期：2026-10-21 | 标签：auto, arxiv, hf-daily-papers
> 核心摘要：研究提出通用双臂灵巧跟踪框架DONIC及包含1575序列的BiHOI数据集。通过PPO与行为克隆反馈闭环及物体中心奖励机制，单策略在未见轨迹上实现86.79%成功率，大幅超越现有双臂基线模型。

# DONIC双臂灵巧跟踪策略发布，零样本泛化成功率达86.79%

::: v-pre
> 本文由 AI 根据公开内容整理，可能有误，请以原文为准：[DONIC: Generalist Dexterous Bimanual Tracking Control from Human Demonstrations](https://arxiv.org/abs/2610.29997)（HF Daily Papers · arXiv 2610.29997）

## 核心内容

针对双臂灵巧操作数据稀缺与接触协调难题，研究团队发布首个通用双臂跟踪器DONIC及配套多视角数据集BiHOI。BiHOI涵盖229种物体共1575个序列，重点记录持续双臂交互并将人类动作重定向为机器人参考轨迹。实验表明，无需微调的单检查点模型在训练物体的未见轨迹上达到86.79%成功率，在完全未见物体上达57.32%，分别超越SOTA基线77.36和45.73个百分点。

DONIC采用近端策略优化（PPO）与行为克隆（BC）构建反馈闭环，将验证成功的执行自动回收为训练演示以解决数据匮乏问题。算法设计物体中心的单手奖励函数，在维持物体运动反馈的同时根据参考接触角色监督各手；引入Clip-Centered Advantages消除参考依赖的优势偏移并保留轨迹内差异。该架构使单一策略能在异构参考下学习协调操作，有效解决了双臂接触丰富任务中的泛化瓶颈。

## 关键看点

- **BiHOI数据集**：包含1575序列、229物体，专注持续双臂交互，支持人机动作为机器人参考重定向
- **训练闭环机制**：结合PPO与行为克隆，自动回收验证成功的执行轨迹作为新训练演示数据
- **奖励与优势设计**：采用物体中心单手奖励与Clip-Centered Advantages，解耦参考依赖并保留轨迹差异
- **泛化性能指标**：单策略零样本在未见轨迹成功率86.79%，未见物体57.32%，显著优于SOTA

## 工程与产业观察

DONIC通过数据回收闭环与精细化奖励设计，突破了双臂灵巧操作对大规模高质量数据的依赖。其单策略跨异构参考泛化能力，为具身智能从仿真到真机的双臂协同提供了可复现的工程范式，验证了在接触丰富任务中算法架构创新比单纯堆砌数据更具边际效益。

## 英文摘要 (Abstract)

Despite recent progress in dexterous reference tracking, existing models focus primarily on single-hand manipulation and a limited range of interactions, leaving the bimanual dexterous behaviors common in real life unaddressed. The scarcity of contact-rich bimanual manipulation data and the increased complexity of coordinating contacts make developing a generalist bimanual tracker particularly challenging. Therefore, we present 1) DONIC, the first generalist bimanual tracker jointly trained across operation categories and human-motion sources, together with 2) BiHOI, a multi-view dataset of 1,575 sequences spanning 229 objects and emphasizing sustained bimanual interaction, where reconstructed interactions are retargeted into robot references. DONIC combines proximal policy optimization with behavioral cloning in a feedback loop that recycles verified successful executions into training demonstrations. An object-centered per-hand reward maintains object-motion feedback while supervising each hand according to its reference contact role. Clip-Centered Advantages remove reference-dependent advantage offsets while preserving within-trajectory differences. Together, these designs enable one single policy to learn coordinated manipulation across heterogeneous references. Without finetuning, a single checkpoint achieves 86.79% success on unseen trajectories of training objects and 57.32% on held-out objects, surpassing the state-of-the-art bimanual tracking baselines by 77.36 and 45.73 percentage points.

Hugging Face Daily Papers 热度：35 upvotes
作者：Mingze Sun, Ziang Cao, Yuan Fei, Qiankuan Tao, Zihan Wang, Tianqi Liu, Junhao Chen, Haiwen Diao, Fangzhou Hong, Zhaoxi Chen, Ziwei Liu
:::

## 原始参考来源 (Sources)

- [HF Daily Papers · arXiv 2610.29997：DONIC: Generalist Dexterous Bimanual Tracking Control from Human Demonstrations](https://arxiv.org/abs/2610.29997)