Texture Classification of a Miniature Whisker Sensor with Varied Contact Pose

Tactile perception using whisker sensor is widely applied to robots under dark and narrow environments. However, most of the existing whisker sensors are relatively large. Meanwhile, most experiments of texture classification using whisker sensors are carried under relatively constrained conditions. In this paper, we developed an ultra-small whisker sensor consisting of a sensing unit and a nylon whisker (3 cm in length) on a circular PCB with diameter of 1.5 cm. The sensor transforms the deflection of whisker into voltage by Wheatstone bridge. In the experiment, the whisker sensor was controlled to contact the surface of four different materials with varied contact pose (different distances and contact angles). The collected data were classified with SVM corresponding to different contact distances and contact angles. The results show that a larger distance and a smaller angle have impact on the amplitude of whisker vibration, resulting in low accuracy. Furthermore, we proposed a thresholding method to confirm the starting point of contact and extract the steady output signal automatically. Eventually, the classification process can be finished within 1 s after contact and the mean classified accuracy is 88.3% for different contact distances, and 85.2% for different contact angles.