ML-based Classification of Roughness Touch Feeling by using a Fiber Bragg Grating Sensor

Abstract

Flexible tactile sensors are important for robots that are used in environments with humans, as they have little physical impact when they contact with people or objects. Most conventional flexible tactile sensors, such as PVDF or strain gages, are based on electrical characteristics. However, these sensors are susceptible to electromagnetic noise, which may reduce the accuracy of measurement. Herein, we developed and evaluated the performance of a tactile sensor using a Fiber Bragg Grating sensor, which is flexible and highly resistant to electromagnetic noise. To discriminating objects with different roughness, we constructed a tactile sensor with optical fiber in which FBG were introduced and fixed in a semicircular shape, and a measurement system that mimics human tactile motion. Vibrations were measured when tracing the surface of seven different aluminum plates with different roughness. Furthermore, we constructed a multi-layer perceptron classification model based on a neural network and calculated the classification accuracy of the vibration information acquired by the sensor. The constructed model has 500-dimensional input layers, 7-dimensional output layers, and 3 hidden layers, and was confirmed to be able to discriminate samples with different roughness with an accuracy of 95.7%.