Zhanpeng He
Papers
10
Total Citations
404
H-Index
9
About
Zhanpeng He is a robotics and machine learning researcher whose work spans reinforcement learning, robot manipulation, and sim-to-real transfer. His research addresses some of the most fundamental challenges in making robots capable, adaptable, and efficient learners in complex real-world environments. He is perhaps best known for his contribution to Meta-World (2019), a landmark benchmark for multi-task and meta reinforcement learning that has garnered over 280 citations and become a standard evaluation framework in the field. This work directly tackled the narrow task distributions that limited prior meta-RL research, opening new avenues for studying how robots can generalize across diverse skills. Beyond benchmarking, He has made significant contributions to robot hand design and dexterous manipulation, including co-design of hardware and control policies using policy gradient methods, and discovering postural synergies for high-dimensional manipulation tasks. His work on 3D dynamic scene representations, video-based learning for long-horizon tasks, and task-aware grasp estimation further demonstrates a commitment to building perception and planning systems that reflect the complexity of real environments. More recently, his research on uncertainty-aware human-in-the-loop decision making highlights a growing interest in safe, collaborative autonomy—an increasingly vital frontier in applied robotics.
Research Focus
Key Achievements
Top Papers
- 1
- 2Co-designing hardware and control for robot hands23 citations · 2021
- 3Learning 3D Dynamic Scene Representations for Robot Manipulation21 citations · 2020
- 4
- 5
- 6
- 7
- 8
- 9
- 10