Newest papers

DONIC: Generalist Dexterous Bimanual Tracking Control from Human Demonstrations

Mingze Sun, Ziang Cao, Yuan Fei, Qiankuan Tao, Zihan Wang, Tianqi Liu, Junhao Chen, Haiwen Diao, Fangzhou Hong, Zhaoxi Chen, Ziwei Liu · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.29997 · cited by 0 works (OpenAlex) · open-access PDF

Korean

단일 객체 참조 추적에 대한 최근의 진보에도 불구하고, 기존 모델은 주로 단일 손 조작과 제한된 범위의 상호 작용에 초점을 맞추고 실제 생활에서 일반적인 두 개의 개체 참조 행동을 해결하지 못하게합니다. 접촉이 풍부한 두 개의 참조 조작 데이터의 부족과 229 개체에 걸친 조정 접촉의 증가 된 복잡성은 특히 도전적입니다. 따라서 우리는 1) DONIC, 최초의 일반적인 두 개의 참조 추적자가 작동 범주와 인간 운동 소스와 함께 공동으로 훈련을 받았습니다. 2) BiHOI, 1.575 시퀀스의 다중 시각 데이터 세트 229 개체를 다루고 지속적인 두 개의 참조를 강조하여 재구성 된 상호 작용이 로봇

Machine translation (Meta M2M100); the English original follows.

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.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Computer science, Artificial intelligence, Tracking (education), Generalist and specialist species, Robot, Control (management)

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange