Papers
4
Total Citations
67
H-Index
3
About
Yujun Zeng is a researcher advancing the frontiers of deep reinforcement learning (DRL), with a focus on enabling intelligent agents to navigate and act in complex, real-world environments. His work tackles two critical bottlenecks in modern AI: sample efficiency and the challenge of learning from high-dimensional visual inputs. In his most cited work (57 citations), Zeng proposed a hybrid deep imitation reinforcement learning framework for target-driven visual navigation, dramatically improving both learning speed and navigation performance in indoor scenes. He has further explored how state representations can be extracted from raw vision data to make DRL more data-efficient, and developed a least-squares truncated temporal-difference method for more stable policy evaluation. Addressing the notoriously difficult problem of robotic control with sparse rewards, Zeng introduced a data-efficient DRL approach that scales continuous robotic tasks without requiring dense feedback signals. His research systematically addresses the gap between simulated success and real-world applicability, making him a notable contributor to the next generation of sample-efficient, vision-based reinforcement learning systems.
Research Focus
Key Achievements
Top Papers
- 1
- 2
- 3
- 4