Terrain Classification Based on Machine Learning Using Thin-film Pressure Sensors on Spherical Robots

Abstract

Environmental recognition by sensors mounted on robots is a pivotal area of research within robotics. This study aims to pioneer a terrain sensing system with multiple thin-film pressure sensors on a spherical mobile robot platform. This system is able to measure the reaction force when the robot makes contact with the ground, thereby obtaining the corresponding stiffness data from different terrain materials. Machine learning techniques are applied to distinguish the terrain characteristics with high accuracy since the data is non-linear, and the patterns of different materials are imperceptible for humans to distinguish.