Grasp Control of a Multi-fingered Robot Hand for Cups of Different Sizes using a Visuomotor Integration Model

Herein, we focused on a robot hand control technology based on image recognition. We conducted experiments on a model that calculates the finger joint angle information of the robot hand from an image of a target object. A model of human visuomotor integration and translation has been proposed, and this model consists of modular auto-encoders corresponding to visual and motor information. To verify the effectiveness of this model, we conducted learning experiments with a real four-fingered robot hand with 16 degrees of freedom. We used cups with different bottom diameters and handle lengths as target objects. We grasped the cups with a robot hand and obtained the finger joint angle when grasping the sides of the cup and when holding the top of the cup as motor information. We conducted a learning experiment using depth images as visual information. As a result, we confirmed that the information on the diameter of the cups could be extracted. Even when images with different cup sizes and handle directions, which were excluded in the training data, were input as test data, the system could represent the cup size. In the hand posture calculation experiment, we confirmed that the posture of the robot hand changes according to the size of these untrained cups.