A Heterogeneous Ensemble Approach for Terrain Classification with Tracked Mobile Robots

Abstract

Terrain classification plays a vital role in the autonomous driving of tracked mobile robots. This article proposes a heterogeneous ensemble method to enhance the classification performance with weak classifiers. Combined with principal component analysis (PCA) and ReiefF, a feature selection approach aims to find an effective feature set. Based on the proposed method, the framework has been developed to simplify the signal process in feature extraction. Extreme Learning Machine (ELM) is utilized as the individual classifier for three kinds of signal sources. With the learning process experience, ensemble weights have been generated to support the whole ensemble flow, which is closely correlated with terrain categories, classifiers and signal sources. Due to the structure of tracked mobile robots, the application maneuvers are extended to the front and rear drive modes, while the steady-state turning and straight-line maneuvers. The results demonstrate that the proposed method can significantly enhance the performance of terrain classification.