Hands Group

Research Overview

In this study, we aim to realize dexterous manipulation by equipping a multi-fingered robot hand, modeled after the human hand, with a large number of tactile sensors and utilizing deep learning. Humans perform various tasks using their five fingers (multi-fingered) and the sense of touch perceived across the entire hand. Enabling robots to perform similar tasks constitutes not only an academic challenge of generating human-like manipulation but also an important effort toward addressing social needs with robot hands that closely resemble human hands.
In particular, much of the research in robotic manipulation has emphasized visual information, which faces challenges such as occlusion and limitations in performing delicate and dexterous manipulation. To achieve more sophisticated manipulation, our approach seeks to leverage bodily information such as tactile and proprioceptive feedback, combining both software aspects centered on deep learning–based control and hardware aspects that emphasize the structure and functions of the human hand.
Through this approach, we aim to achieve challenging multi-fingered hand tasks in which multiple fingers work in coordination to skillfully manipulate difficult objects—such as soft or slippery items—within cluttered and constrained environments.