I am a PhD student at the School of Computing and Data Science, The University of Hong Kong, advised by Prof. Jia Pan in the PANDA Lab.
My research focuses on robot learning for manipulation, especially imitation learning and reinforcement learning. I am interested in improving learned robot policies through RL post-training.
I interned at Shanghai Qizhi Institute, working on RL post-training for imitation policies, and at Xiaomi Robotics in 2025, working on teleoperation with ViTai tactile sensors.
I received my M.S. (2023) and B.S. (2020) in Mechanical Engineering from Tsinghua University, advised by Prof. Shengbo Li. I also collaborated with Prof. Jianyu Chen on safe RL and optimal control.
Email: djyu [at] connect [dot] hku [dot] hk
IEEE T-RL, 2026 (accepted, formal version in progress)
ZPRL steers pretrained robot policies by perturbing bottleneck latents, improving sample efficiency while producing smoother actions.
Science Robotics, 2026
project page / paper / code
RL-100 unifies imitation learning and real-world RL for 3D diffusion policies, enabling reliable and efficient robotic manipulation.
IEEE T-ASE, 2025
Vision-language models identify contact modes to coordinate bimanual keyposes for hierarchical robot manipulation.
WAFR, 2024
A hierarchical bimanual policy that reduces accumulated errors and inference latency with keyposes and consistency models.
IEEE T-ASE, 2023
A distributional reachability certificate accounts for dynamics uncertainty to reduce safety violations during exploration.
ICML, 2022
paper / code / environments
Hamilton–Jacobi reachability characterizes persistently safe states for constrained reinforcement learning.
Reviewer
Member, Tech Service Team of the Work-Study Program
Tsinghua University, 2020–2023
Excellent Team Member of the Month (x3) and of the Semester (x2); Outstanding Individual of the Work-Study Program in 2021 (top 10% in the university).
Teaching Assistant, The University of Hong Kong
