Servo load analysis for the classification of surface of locomotion in a modular snake-like robot

Abstract

This paper shows the possibility of classifying the surface of locomotion of a modular snake-like robot only from torque and current sensors in the servo-motors. Locomotion in modular snake-like robots is made from gaits that involve the entire body structure, in this particular work we use a modular snake-like robot consisting of 16 modules located 90 degrees rotated one with respect to the previous, this robot is built from Dynamixel AX-12 servos, these servos provide load information based on torque and power consumption. This article presents an analysis from two classifiers, supervised and unsupervised for load temporary data in each of the 16 modules, for three different gaits used in the robot, linear progression, side winding and lateral rolling to make an identification of the characteristics of the surface on which the robot is moving, all without the need for other external sensors. The reported results are obtained by applying two classification techniques, the first supervised (SVM) and the second unsupervised (K-means). Is concluded that it is possible to make a classification between surfaces, knowing previously the selected gait, reaching even to 100% accurancy for certain gaits.