Papers
6
Total Citations
150
H-Index
5
About
Kuo-Hao Zeng is a researcher working at the intersection of computer vision, robotics, and embodied artificial intelligence, with particular expertise in visual perception and autonomous navigation. His early and most-cited work introduced the Omnidirectional Convolutional Neural Network (O-CNN), a novel architecture designed to tackle visual place recognition under severe camera pose variation using omnidirectional cameras — a contribution that has garnered over 74 citations and demonstrated meaningful advances in robust spatial understanding for navigation systems. Zeng also contributed to AllenAct, an influential framework for Embodied AI research that has helped unify deep reinforcement learning experiments across computer vision, NLP, and robotics communities, accumulating 44 citations since its 2020 release. His research portfolio extends into video captioning with spatio-temporal attention mechanisms and visual common sense reasoning, exploring what objects are meaningfully absent from a scene. Most recently, his work on FLaRe addresses a critical gap in generalist robot policies by applying large-scale reinforcement learning fine-tuning to improve adaptability beyond behavior cloning limitations. Across his career, Zeng has consistently pushed the boundaries of how machines perceive, reason about, and act within complex visual environments.
Research Focus
Key Achievements
Top Papers
- 1Omnidirectional CNN for Visual Place Recognition and Navigation74 citations · 2018
- 2AllenAct: A Framework for Embodied AI Research44 citations · 2020
- 3Omnidirectional CNN for Visual Place Recognition and Navigation13 citations · 2018
- 4Video Captioning via Sentence Augmentation and Spatio-Temporal Attention10 citations · 2017
- 5
- 6Seeing the Unseen: Visual Common Sense for Semantic Placement3 citations · 2024