Rover odometry aided by a star tracker

This paper develops a practical framework for estimating rover position in full-dark conditions by correcting relative odometric estimates with periodic, absolute-attitude measurements from a star tracker. The framework is validated using just under 2.5 kilometres of field data gathered at the University of Toronto´s Koffler Scientific Reserve at Jokers Hill (KSR) comprised of both wheel odometry and lidar-based Visual Odometry (VO). It is shown that for the wheel odometry solution, the final estimate of rover position was within 21 metres of the groundtruth as calculated by a differential GPS receiver, or 0.85% of the total traverse distance. When the star tracker measurements are artificially limited to occurring approximately every 250 metres, the algorithm still performs well, giving a final position error of 75.8 metres or 3.0%. Preliminary results to replace wheel odometry with lidar-based VO for the development a full-dark visual solution are also presented. The lidar-based VO solution is shown to be capable of outperforming wheel odometry, but more work is required to develop methods to handle the variety of terrain conditions encountered.