Terrain identification methods for planetary exploration rovers

Abstract

Autonomous mobility in rough terrain is becoming increasingly important for planetary exploration rovers. Increased knowledge of local terrain properties is critical to ensure a rover's safety, especially when driving on slopes or rough surfaces. This thesis presents two methods for using on-board sensors to identify local terrain conditions. The first method visually measures sinkage of a rover wheel into deformable terrain, based on a single color or grayscale image from a camera with a view of the wheel- terrain interface. Grayscale intensity is computed along the rim of the wheel, and the wheel-terrain interface is identified as the location with maximum change in intensity. The algorithm has been shown experimentally to give accurate results in identifying the terrain characteristics under a wide range of conditions. The second method classifies terrain based on vibrations induced in the rover structure by rover-terrain interaction during driving. Vibrations are measured using an accelerometer on the rover structure. The method uses a supervised learning approach to train a classifier to recognize terrain based on representative vibration signals during an off-line learning phase. Real-time terrain classification uses linear discriminant analysis in the frequency domain to identify gross terrain classes such as sand, gravel, or clay. The algorithm is experimentally validated on a laboratory testbed and on a rover in outdoor conditions. Results demonstrate the robustness of the algorithm on both systems.