Papers
35
Total Citations
429
H-Index
11
About
Huazhe Xu is a robotics and machine learning researcher whose work spans reinforcement learning, tactile sensing, and robot manipulation. His research tackles some of the field's most demanding challenges: enabling robots to interact intelligently with complex physical environments through perception, learning, and dexterous control. Xu's contributions to multi-task reinforcement learning include the development of Soft Modularization, which elegantly addresses how neural networks can share and reuse parameters across diverse tasks. His pioneering work on vision-based tactile sensors — particularly the 9DTact and DTact systems — has advanced robots' ability to reconstruct 3D contact geometry and estimate forces with remarkable precision, each garnering nearly 60 citations. In deformable object manipulation, his RoboCraft framework demonstrated how graph neural networks can help robots model and shape elasto-plastic materials, opening pathways to complex real-world tasks like food preparation. His cross-modal Transformer approach to quadrupedal locomotion showcases a broader interest in embodied intelligence across diverse robotic platforms. More recently, work on affordance generalization and diffusion-based reward learning reflects a forward-looking engagement with semantic understanding and scalable robot learning. With over 300 cumulative citations across a focused and rapidly growing body of work, Xu represents an influential voice in next-generation robot intelligence.
Research Focus
Key Achievements
Top Papers
- 1
- 2Multi-Task Reinforcement Learning with Soft Modularization58 citations · 2020
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10Diffusion Reward: Learning Rewards via Conditional Video Diffusion12 citations · 2024