Thin film object recognition system using a two-legged piezoelectric actuator-sensor pair with machine learning

Abstract

Object recognition is important for robots to sense the surrounding environment and manipulate tools in the real world. Recently, many advanced object recognition systems using vision sensors have been reported. However, in vision-sensing systems, counting the number of sheets in very thin objects is difficult. Moreover, various transparent objects can be recognized as the same object. To address these issues, we introduce a thin-film object recognition method using a two-legged piezoelectric actuator-sensor pair. Specifically, it consists of an actuator and a sensor made of polyvinylidene fluoride with a piezoelectric effect. When a high-frequency voltage sweep is applied to the actuator, it generates vibration, and the sensor measures the frequency response through the object. The sensing data are trained using a long short-term memory network. Using the trained model, we demonstrate two types of recognition: counting the number of sheets of paper and classifying materials made of transparent thin films, and both experimental results show significantly high accuracy.