Offroad Terrain Classification for Mobile Robots

In recent years, the capabilities of mobile robots have increased significantly, opening up new potential applications in a variety of settings. One area where mobile robots show great promise is in offroad terrain classification, where the ability to accurately identify and navigate different types of terrain is critical. In this paper we present a new pipeline for terrain classification in offroad environments. The main contributions of this pipeline are a combined offroad dataset utilizing publicly available datasets and minimizing manual labeling, a validated network architecture for optimized generalization to new environments, and a post-processing step to improve the reliability of the classification in the context of offroad navigation. The proposed approach was evaluated using publicly available data as well as newly collected data from offroad environments.