Skip to Main Content
The work presented here empirically analyzes the design of the tightly-coupled position, velocity, and attitude estimator used as a feedback signal for autonomous navigation in a large scale robot driving in urban settings. The estimator fuses GNSS/INS signals in an extended square root information filter (ESRIF), a numerically-robust implementation of an extended Kalman filter (EKF), and was used as the basis for Cornell University's 2007 DARPA Urban Challenge robot, "Skynet." A statistical sensitivity analysis is conducted on Skynet's estimator by examining the changes in its behavior as critical design elements are removed. The effects of five design elements are considered: map aiding via computer vision algorithms, inclusion of differential corrections, filter integrity monitoring, Wide Area Augmentation System (WAAS) augmentation, and inclusion of carrier phases; the effects of extensive signal blackouts are also considered. Metrics of comparison include the statistical differences between the full solution and variant; the Kullback-Leibler divergence; and the average and standard deviation of the position errors, attitude errors, and filter update discontinuities.