Terrain classification for mobile robots traveling at various speeds: An eigenspace manifold approach

Abstract

Unmanned ground vehicles (UGV's) commonly used in military applications must possess the capability to traverse various terrains that may largely affect the performance and controllability of the vehicle. A UGV that can autonomously perceive its terrain using navigational sensors can make necessary changes to its control strategy. The research presented uses the output of the induced vehicle's vibration measured by navigational sensors to classify the underlying terrain at multiple speeds. The classification algorithm incorporates Principal Component Analysis (PCA) for feature extraction and dimension reduction. The PCA transformation coefficients are then used to develop a manifold curve that uses these known coefficients to interpolate unknown coefficients of the terrains as the robot's speed changes. Experimental data is collected using two distinctly different unmanned ground vehicle platforms. Results demonstrate the performance of the method for classifying multi-differentiated terrains broadly classified as grass, asphalt, mud, and gravel.