Yasuto Yokota
Papers
2
Total Citations
24
H-Index
2
About
Yasuto Yokota is a robotics researcher focused on advancing autonomous object manipulation through deep learning. His primary research areas include self-supervised learning for robotic grasping, multi-task learning, and few-shot classification—all aimed at reducing the costly data requirements that limit industrial robot deployment. Yokota’s most impactful contribution is his 2020 work on online self-supervised learning for object picking, which introduced a metric learning approach that enables robots to detect optimum grasping positions without complete ground-truth labels. This method, cited 16 times, addresses a critical bottleneck: the inability of trial samples to provide full observable feedback during autonomous training. In a second key paper (8 citations), Yokota proposed a multi-task learning framework that simultaneously performs grasping-position detection and few-shot classification, drastically reducing the labeled images needed when factories frequently change product shapes. His work directly tackles the practical challenge of retraining costs in manufacturing, offering scalable solutions for picking robots. By combining self-supervision with multi-task architectures, Yokota has made notable strides toward more adaptable, data-efficient robotic systems—a vital step for flexible automation in real-world industrial settings.
Research Focus
Key Achievements
Top Papers
- 1
- 2