Research

My research focuses on world models, robot foundation models, and learning from human and robot demonstrations.

Selected Publications

* Equal contribution. My name is shown in bold.

Liber-0: Emergent Zero-Shot Capabilities in Multi-Step Manipulation from 2.2 Million-Hour Human Experience

LiberAI Team
LiberAI Tech Report
[Technical Report]

World action modeling for zero-shot, multi-step robot manipulation through learning from human experience.

RDT2: Exploring the Scaling Limit of UMI Data Towards Zero-Shot Cross-Embodiment Generalization

S. Liu*, B. Li*, K. Ma*, L. Wu*, H. Tan, X. Ouyang, H. Su, J. Zhu.
ICML 2026 ยท Co-first author
[Paper] [Project] [Code]

Scaling handheld UMI data to a 7B vision-language-action model trained with over 10,000 hours of demonstrations, studying generalization to unseen objects, scenes, instructions, and robot embodiments.

RDT-1B: A Diffusion Foundation Model for Bimanual Manipulation

S. Liu*, L. Wu*, B. Li, H. Tan, H. Chen, Z. Wang, K. Xu, H. Su, J. Zhu.
ICLR 2025
[Paper] [Project] [Code]

A 1.2B-parameter diffusion foundation model for bimanual manipulation, combining heterogeneous robot pretraining with a unified action representation and task-specific fine-tuning.

GEM: Generative Supervision Helps Embodied Intelligence

R. Zhao, B. Li, Z. Liu, Y. Liang, J. Ye, F. Liu, D. Wu, Z. Wang, X. Yu, Y. Rao, H. Hu, J. Zhu.
ECCV 2026
[Paper]

Depth-generation supervision for embodied vision-language pretraining and downstream manipulation through GEM-VLA.

An Atomic Skill Library Construction Method for Data-Efficient Embodied Manipulation

D. Li, B. Peng, C. Li, N. Qiao, Q. Zheng, L. Sun, Y. Qin, B. Li, et al.
arXiv preprint, 2025
[Paper]