An Optical Tactile Sensor With Structural Color Using Deep Learning Method

Touch is an indispensable perception means for robots. However, tactile sensors on robots generally have insufficient resolution and flexibility. In recent years, optical tactile sensors have attracted more and more attention due to its high resolution. We designed a small optical tactile sensor based on the principle of diffraction, it breaks through the limitation of marker density on resolution. The sensor mainly includes a LED light source, a flexible grating film and a miniature RGB camera. Irregular light source causes irregular diffraction patterns, which makes data processing difficult. To address this problem, a deep learning algorithm is used to process diffraction patterns, and then a series of evaluations are made on the effect of the algorithm and the accuracy of the sensor. In addition, we compare the effect of the parallel light source with the common LED light source on sensor accuracy. The results show that our sensor has no special requirement for the parallelism of the light source.