Classification of Floor Materials Using Piezoelectric Actuator-Sensor Pair and Deep Learning for Mobile Robots

Analyzing floor surface materials is critical for controlling the motion and tasks of mobile robots. In this study, we propose a novel method for classifying floor materials for indoor mobile robots using a piezoelectric actuator-sensor pair and deep learning. This method can classify the floor properties itself with isolated sensing system while the mobile robot is moving. The piezoelectric pair is a thin-film type. It consists of an actuator and a sensor. The sensing pair is positioned at the bottom of the robot. When the robot moves forward, the sensing part collects the electrical responses from the actuator. Since one-dimensional data is collected through the piezoelectric actuator-sensor pair, the size of the system is small and the data processing speed can be reduced. Using this mechanism, experiments were conducted to classify various materials of floor surfaces in indoor environments. The sensing data were processed by fast Fourier transform, high-pass filter, polynomial fitting, and sampling to be used as inputs for machine learning of the classification model. Specifically, the trained model achieved a high accuracy of 95.4%. In addition, the training data were verified using the k-means clustering method. Moreover, the effect of the physical properties on the sensor data was analyzed to investigate the relationship between the materials and the sensing outputs.