Real-time visual sinkage detection for planetary rovers

Abstract

Identifying wheel sinkage for planetary exploration rovers can give a critical insight about the terrain traversability and in particular the characteristics of deformable soils that the rover travels on. This paper presents a monocular vision based approach that can detect and estimate the sinkage of a hybrid legged wheel in real time and robustly, with little sensitivity to changing operational conditions. The proposed method involves color-space segmentation that identifies the leg contour and consequently depth of wheel sinkage into the regolith. In addition, it enables dynamic analysis of the sinkage, hence making detection of non-geometric hazards possible while the rover moves. Extensive field trials have been conducted on natural deformable terrain. The experimental results demonstrate that the average discrepancy against annotated images is less than 1%.