Papers
2
Total Citations
57
H-Index
2
About
Haoyi Zhu is a rising researcher in robotics and artificial intelligence, whose work focuses on enabling robots to learn and generalize diverse manipulation skills across open domains. His primary research areas include one-shot imitation learning, robotic foundation models, and multi-domain policy adaptation. Zhu’s most impactful contribution is the RH20T dataset (2024, 55 citations), a comprehensive robotic dataset designed to facilitate learning diverse skills from a single demonstration, addressing a critical bottleneck in open-world robotic manipulation. This work has become a key resource for advancing imitation learning and foundation model research in robotics. More recently, Zhu introduced Tra-MoE (2025), a trajectory prediction model that leverages out-of-domain data to improve generalization and adaptive policy conditioning across multiple domains. By tackling the challenge of learning from broad, heterogeneous data sources, Zhu’s research pushes toward more robust and versatile robotic systems. His contributions are particularly notable for bridging dataset construction with algorithmic innovation, providing both the data and the models needed to scale robot learning. As a young researcher, Zhu’s work is already shaping the trajectory of generalizable robotic skill acquisition.
Research Focus
Key Achievements
Top Papers
- 1
- 2