Biography
I am a second-year Computer Science Ph.D. student at
Previously, I obtained my M.Phil. degree in Electronic and Computer Engineering at
Here is my CV/Resume.
Research
Make Robots Work.
I am interested in understanding how intelligent systems can effectively acquire, represent, and apply knowledge about their environments to solve complex problems. My research focuses on (1). developing new machine learning algorithms to enhance the capabilities of robots and (2). building more scalable robot learning systems.
Selected Publications
My past explorations. * indicates equal contribution.
Dexterity Gen: Foundation Controller for Unprecedented Dexterity
Z.H. Yin , C. Wang, L. Pineda, F. Hogan, K. Bodduluri, A. Sharma, P. Lancaster, I. Prasad, M. Kalakrishnan, J. Malik, M. Lambeta, T. Wu, P. Abbeel, M. Mukadam.
Tech Report 2025 PDF Project
Geometric Retargeting: A Principled, Ultrafast Neural Hand Retargeting Algorithm
Z.H. Yin , C. Wang, L. Pineda, K. Bodduluri, T. Wu, P. Abbeel, M. Mukadam.
Tech Report 2025 PDF Project
Offline Imitation Learning through Graph Search and Retrieval
Zhao-Heng Yin , Pieter Abbeel
RSS 2024 PDF Project
Using graph search and retrieval to learn from suboptimal human demonstrations.Twisting Lids off with Two Hands
Toru Lin*, Zhao-Heng Yin* , Haozhi Qi, Pieter Abbeel, Jitendra Malik
CoRL 2024 PDF Project
Bimanual dexterous manipulation through sim-to-real.Rotating without Seeing: Towards In-hand Dexterity through Touch
Zhao-Heng Yin* , Binghao Huang*, Yuzhe Qin, Qifeng Chen, Xiaolong Wang
RSS 2023 PDF Project Code
Dexterous robotic hand with touch-only control.
Spatial Generalization of Visual Imitation Learning with Position-Invariant Regularization
Zhao-Heng Yin , Yang Gao, Qifeng Chen
RSS 2023 Workshop PDF Code
Spatial symmetry as policy regularization.
Planning for Sample Efficient Imitation Learning
Zhao-Heng Yin , Weirui Ye, Qifeng Chen, Yang Gao
NeurIPS 2022 PDF Code
We propose EfficientZero-Continuous and EfficientImitate for sample efficient robot learning.
Cross Domain Robot Imitation with Invariant Representation
Zhao-Heng Yin , Lingfeng Sun, Hengbo Ma, Masayoshi Tomizuka, Wu-Jun Li
ICRA 2022 PDF Code
Learning from a similar yet different expert.
DexPoint: Generalizable Point Cloud Reinforcement Learning for Sim-to-Real Dexterous Manipulation
Yuzhe Qin*, Binghao Huang*, Zhao-Heng Yin , Hao Su, Xiaolong Wang
CoRL 2022 PDF
Diverse Critical Interaction Generation for Planning and Planner Evaluation
Zhao-Heng Yin* , Lingfeng Sun*, Liting Sun, Masayoshi Tomizuka, Wei Zhan
IROS 2021 PDF
TOMA: Topological Map Abstraction for Reinforcement Learning
Zhao-Heng Yin , Wu-Jun Li
arXiv 2020 PDF
Random
Some fun projects I participated in the past.
Bio-inspired* Direct-Drive Palmar Manipulation with Visual Imitation Learning
with Ka Hei Mak and Jungwon Seo (2021)
* guess where the inspiration comes from :P
My Chinese Name (surname first): 殷 兆恒.
Some of my past wild arguments in the lab: "imitation learning solves robotics", "imitation learning is a dead end", "do not touch simulation, that will kill you", "only simulation can save robotics", "we should fall back to classical stuffs", "dexterous hands are not really necessary." Finally, I've found some inner peace and faced the tough truth: robotics is hard and there is no shortcut solution. I just need to figure out the least painful way forward and keep moving.
There are just too many corner tasks (or cases) in robotic manipulation that can defeat any of the existing robot learning pipelines. Now I appreciate human hand a lot. It is so surreal.