Dynamic Tactile Exploration for Texture Classification using a Miniaturized Multi-modal Tactile Sensor and Machine Learning
Bruno Monteiro Rocha Lima, Vinicius Prado da Fonseca, Thiago Eustaquio Alves de Oliveira, Qi Zhu, Emil M. Petriu
Abstract
The findings of this paper enhance the knowledge of machine learning in tactile texture recognition. Synthetic perception of textures was achieved using a recently developed miniaturized multi-modal tactile sensor. Thirteen commonly used textures were dynamically explored and classified using supervised learning classifiers. The best result was achieved by the Extra Trees classifier with an average accuracy score of 95%. Other classifiers also had high-performance results with an average accuracy score of 92% (Random Forest), 91% (Support-Vector Machines), and 88% (Multilayer Perceptron). Among pressure, acceleration and magnetic flux, the sensor's angular velocity proved to be the most efficient feature to classify textures. The results also showed a correct classification between two textures that were originated from the same anisotropic material. This indicated that the exploration in different directions produced distinctive outputs and, therefore, the exploration in two dimensions have to be analyzed in future works.