Traversability estimation system for mobile robot in heterogeneous environment with different underlying surface characteristics

One of the key tasks of Outdoor-type mobile robotics is traversability estimation of underlying surfaces in a in a priori of an unknown heterogeneous environment. The paper presents practical realization of traversability estimation system based on group method of data handling (GMDH). This method is classical technique of data mining and one of the first techniques of Deep Learning. The results of color, geometry and texture features extraction by developed computer vision unit are presented step by step. Also the results of model training (Twice-Multilayered Modified Polynomial Neural Network with active neurons is used as one of the GMDH algorithms) for different input features subsets combinations and for two variants of traversability estimation (the robot leaves the area being traversed, but remains within a specified radius and traversing an area within a given time) are considered. The obtained results testify the efficiency of the developed traversability estimation system.