Study on Simultaneous Classification of Object Properties by Using Soft Tactile Sensor Array System Utilizing Local Spatial-temporal Information of Contact

This paper deals with simultaneous classification of shape, size, and softness of cylindrical objects by using soft tactile sensor array system of a five-links single robotic finger. The front surface of each link is covered with semicircular silicone rubber equipped with 235 small on-off switches. The on-off data from these switches obtained during grasping an object is transformed into a spatial-temporal matrix form. As 8 cells around the contact switch is considered useful to extract the local spatial-temporal character of contact physics, the frequency of the 8 cells patterns composed of 0-1 data around the switch contacted is obtained for each object and then used to form a contact-feature vector. This vector is obtained 10 times of experimental trial, corresponding to each object. The vectors are classified by Mahalanobis distance for 27 objects (3 shapes, 3 sizes and 3 kinds of softness) resulting in 48 grasping postures. By using the most effective 7 kinds of 8-cells patterns, over 95% classification accuracy is obtained.