Shiwei Lian
Papers
2
Total Citations
13
H-Index
2
About
Shiwei Lian is a rising researcher in embodied AI and deep reinforcement learning (DRL), with a focused expertise in object-goal visual navigation. His work directly tackles the critical challenge of generalization—how to train DRL agents that can navigate unfamiliar environments without retraining. Lian’s major contribution is the introduction of **TDANet (Target-Directed Attention Network)**, a novel architecture that achieves zero-shot navigation ability by learning domain-independent visual representations. This work, which has already garnered 8 citations since its 2024 publication, enables agents to recognize and navigate toward target objects even when object classes and placements differ from training scenarios. Complementing this, Lian developed a **transferability metric using scene similarity and local map observation** (5 citations), providing a principled way to quantify and predict how well a DRL navigation policy will perform in a new environment. This metric offers a practical tool for assessing model robustness before deployment. By addressing the fundamental gap between simulated training and real-world testing, Lian’s research is paving the way for more reliable and adaptable autonomous navigation systems, making him a notable emerging voice in the field of intelligent robotics.
Research Focus
Key Achievements
Top Papers
- 1
- 2