ABOUT ME

I have focused on the study of reinforcement learning (RL) since obtaining my bachelor’s degree at the Beijing Institute of Technology. My research is driven by the observation that traditional RL algorithms often struggle with real-world challenges, particularly when faced with sparse feedback and complex action distributions. During my doctoral research, I addressed these challenges by developing novel algorithms aimed at improving both the learning efficiency and stability in environments with limited feedback. My work focuses on refining exploration strategies and utilizing auxiliary tasks to mitigate the negative effects of sparse rewards. Additionally, I am exploring solutions to complex multi-task manipulation problems by leveraging large-scale offline demonstrations. Ultimately, my goal is to develop general manipulation policies that can autonomously recover and improve themselves in dynamic and unpredictable environments.

Previously, I spent two years at Midea Group researching end-to-end robot manipulation policies for service robots using reinforcement learning. Prior to that, I interned at DiDi AI Lab, developing autonomous driving policies with visual reinforcement learning.

🔥 News

• 2025.06: 🎉 Our paper “\(\mathrm{T}^2\)-VLM” has been accepted to ICCV 2025. Many thanks to all co-authors for their excellent contributions. 👉 [Paper]
• 2025.06: 🎉 Our paper “HACTS” has been accepted to IROS 2025. Many thanks to all co-authors for their excellent contributions. 👉 [Paper]
• 2025.06: 🎉 Our team received the Honorable Mention Prize at MOASEI 2025 during the 24th AAMAS! Congratulations to my supervised students Yu Zou, Tianjiao Yi, and Yuxiang Song for their outstanding contributions. 👉 [Award Certificate] [Website]