Comparative analysis of classification algorithms on tactile sensors

Abstract

A comparative analysis of classification algorithms of iCub platform humanoid hand tactile sensors is presented. The experimental data were analyzed with different learning supervised classification algorithms: Decision Trees Classifiers, k-Nearest Neighbors Classifiers (kNN), and Support Vector Machines (SVM). The best result was obtained with a Gaussian SVM kernel, which allowed 97.4% accuracy using 20% data for holdout validation. The results indicate the potential of categorization and learning of robotic hands for object grasping and manipulation.