Due to the absence of any satellite-based global positioning system on Mars, the Mars Exploration Rovers commonly track position changes of the vehicle using a technique called visual odometry (VO), where updated rover poses are determined by tracking keypoints between stereo image pairs. Unfortunately, the error of VO grows super-linearly with the distance traveled, primarily due to the contribution of orientation error. This thesis outlines a novel approach incorporating sun sensor and inclinometer measurements directly into the VO pipeline, utilizing absolute orientation information to reduce the error growth of the motion estimate. These additional measurements have very low computation, power, and mass requirements, providing a localization improvement at nearly negligible cost. The mathematical formulation of this approach is described in detail, and extensive results are presented from experimental trials utilizing data collected during a 10 kilometre traversal of a Mars analogue site on Devon Island in the Canadian High Arctic.
展开▼
