Yuzi Kanazawa

Papers

Total Citations

H-Index

About

Yuzi Kanazawa is a researcher at the forefront of robotic manipulation and computer vision, specializing in self-supervised and few-shot learning for industrial automation. Her work addresses a critical bottleneck in deploying deep learning for picking robots: the prohibitive cost of collecting and labeling vast training datasets when object shapes frequently change in factory settings. Kanazawa’s major contribution is pioneering online self-supervised learning frameworks that allow robots to autonomously collect and learn from their own trial data, eliminating the need for complete ground-truth labels. Her most cited paper, "Online Self-Supervised Learning for Object Picking" (2020, 16 citations), introduced a metric learning approach to detect optimal grasping positions, demonstrating how robots can improve their performance during operation. Building on this, her "Multi-task Learning Framework for Grasping-Position Detection and Few-Shot Classification" (2020, 8 citations) further advanced the field by enabling models to generalize from just a handful of labeled examples. Kanazawa’s work is notable for its practical impact on reducing retraining costs in dynamic manufacturing environments, making deep learning-based robotic picking more scalable and economically viable. Her research bridges the gap between theoretical self-supervised learning and real-world industrial robotics.