Papers
6
Total Citations
69
H-Index
4
About
Kaiyong Zhao is a computer vision and robotics researcher whose work centers on stereo vision, depth estimation, and scene understanding for indoor robotic systems. He is perhaps best known for developing the IRS (Indoor Robotics Stereo) dataset, a large-scale resource designed to train deep learning models for disparity and surface normal estimation — work that has accumulated over 57 citations across multiple publications and addresses a critical gap in training data for robotic perception. By emphasizing stereo over monocular vision, Zhao's research delivers more geometrically accurate scene reconstructions essential for robotic localization, navigation, and interaction. Beyond dataset creation, Zhao has contributed a comprehensive review of vision-based robotic grasp detection, synthesizing advances across object localization, pose estimation, and grasp estimation into a unified framework that serves as a valuable reference for robotics practitioners. More recently, he has expanded his research frontier into panoramic depth estimation, tackling the distinctive challenges of spherical image distortion and discontinuity through novel approaches such as SphereDepth and SphereFusion. Collectively, his body of work bridges the gap between deep learning methodology and real-world robotic applications, making him a notable contributor to the field of embodied AI perception.
Research Focus
Key Achievements
Top Papers
- 1
- 2
- 3
- 4
- 5SphereDepth: Panorama Depth Estimation from Spherical Domain3 citations · 2022
- 6SphereFusion: Efficient Panorama Depth Estimation via Gated Fusion2 citations · 2025