INCREMENTAL MARS ROVER LOCALIZATION USING DESCENT AND ROVER IMAGERY

This paper presents the result of field tests of incremental Mars rover localization using descent and rover imagery, conducted at Silver Lake, CA, in April 1999 and May 2000. The simulated Mars descent images were acquired by a camera system on a helicopter. In addition to the FIDO rover images acquired in 1999, a pair of digital cameras simulated rover navigation cameras and took forward, backward, and side-looking images on 18 stations of a traverse extending about 1 km from the "landing center". At each station, the tilt angle was adjusted to ensure an overlap between the forward and backward rover images. An incremental bundle adjustment model is developed which performs rover localization without repeating computations involving all images. To do this, the corresponding features between the descent and rover images and features between forward and backward rover images are detected and serve as tie points for the incremental bundle adjustment. The result of the incremental bundle adjustment is compared with that of an integrated adjustment using measurements of all descent and rover images together. The result is also compared with the ground truth of GPS data. An accuracy assessment is performed to make sure that the rover positions are located within an accuracy of 1 to 4.5 meters of location error for a distance of 1 km from the landing center.