Xiaodan Liang
Papers
14
Total Citations
611
H-Index
9
About
Xiaodan Liang is a prominent researcher whose work spans computer vision, embodied artificial intelligence, and human-robot interaction. Beginning with foundational contributions in RGB-D scene understanding — most notably the LSTM-CF framework (2016, 188 citations), which unified context modeling and fusion for depth-aware scene labeling — Liang progressively expanded into the ambitious frontier of embodied AI, where intelligent agents must perceive, reason, and act within complex three-dimensional environments. A defining thread throughout Liang's career is enabling machines to understand and anticipate human behavior. Her 2018 work on teaching robots to predict human motion (125 citations) demonstrated how deep learning could equip robots with proactive collaborative capabilities. This evolved into landmark contributions in vision-and-language navigation, including the SOON benchmark (115 citations) and NavCoT, which leverages large language models for disentangled navigational reasoning. Her 2025 comprehensive survey on embodied AI (77 citations) signals her role as a field synthesizer, bridging cyberspace and physical-world intelligence. Liang's body of work reflects a coherent vision: building robots that see, communicate, predict, and navigate with human-like fluency — making her an essential voice in the ongoing pursuit of artificial general intelligence.
Research Focus
Key Achievements
Top Papers
- 1
- 2Teaching Robots to Predict Human Motion125 citations · 2018
- 3SOON: Scenario Oriented Object Navigation with Graph-based Exploration115 citations · 2021
- 4
- 5
- 6Self-Motivated Communication Agent for Real-World Vision-Dialog Navigation22 citations · 2021
- 7Deep Learning for Embodied Vision Navigation: A Survey13 citations · 2021
- 8RGB-D Scene Labeling with Long Short-Term Memorized Fusion Model.12 citations · 2016
- 9
- 10