Vibrotactile Recognition of Surface Textures by a Humanoid Robot

Fig. 1: The robot, shown here scratching one of the 20 surfaces used in the experiments. This study investigates the use of a vibrotactile sense for surface texture recognition by a humanoid robot. The sensor is an artificial fingernail with an attached 3-axis accelerometer, which the robot uses to scratch different surfaces. Our method combines frequency-domain analysis of the acceleration measurements with the Support Vector Machine (SVM) learning algorithm to recognize surfaces. Using this approach the robot was able to recognize twenty different surfaces with accuracy significantly better than chance. The experimental results also show that combining predictions from multiple different scratches on a test surface results in higher recognition accuracy than any isolated scratch alone.