Kai Xu

National University of Defense Technology

Papers

Total Citations

H-Index

About

Kai Xu is a computer vision and robotics researcher whose work bridges fundamental perception problems with practical real-world applications. His research spans two interconnected domains: visual correspondence and template matching, and intelligent robotic manipulation through reinforcement learning. In his 2024 work on differentiable coarse-to-fine correspondence refinement, Xu tackled a long-standing challenge in template matching — a cornerstone task in manufacturing and robotic pose estimation. By introducing a learnable, differentiable refinement pipeline, his approach achieves accurate correspondence even under difficult conditions where traditional methods struggle, garnering 17 citations since publication and signaling meaningful traction in the computer vision community. Complementing this, his 2021 contribution on online 3D bin packing demonstrated a reinforcement learning framework capable of generating practically feasible packing policies without foreknowledge of incoming item sequences — a constraint that makes the problem significantly harder and more industrially relevant. This work reflects Xu's consistent focus on closing the gap between theoretical algorithms and deployable robotic systems. Across both contributions, Xu's research is characterized by a commitment to solving problems at the intersection of perception and physical manipulation, making his work particularly valuable for students and researchers working in industrial robotics, autonomous systems, and embodied AI.