Object shape discrimination using sensorized glove

The sense of touch is crucial for distinguishing different shapes like boxes, cylinders and spheres. This paper proposes a machine learning based model to distinguish between shapes using kinesthetic information from the joint angles of the fingers. The training data from a sensorized glove was used to train a multi-layer support vector machine with a radial basis kernel. When used on simple shapes the proposed model obtained 100% accuracy. The accuracy dropped down to 71% when it was trained with shapes held in more than one way. The joints of the fingers that were critical in holding a particular shape was also identified in this paper.