**4.2 GNSS/IMU sensor fusion scheme**

This urban navigation is based on detecting and mitigating GNSS errors caused by condensed high buildings interfering signals going through [5]. It is using a map-aided adaptive fusion scheme. The method estimates the current active map segment using dead-reckoning and robust map-matching algorithms modeling the vehicle state history, road geometry, and map topology in a hidden Markov model

(HMM). The Viterbi algorithm decodes the HMM model and selects the most likely map segment. The projection of vehicle states onto the map segment is used as a supplementary position update to the integration filter. The solution framework has been developed and tested on a land-based vehicular platform. The results show a reliably mitigate biased GNSS position and accurate map segment selection in complex intersections, forks, and joins. In contrast to common existing adaptive Kalman filter methods, this solution does not depend on redundant pseudo-ranges and residuals, which makes it suitable for use with arbitrary noise characteristics and varied integration schemes.

### **4.3 Navigation based on compass-based navigation control law**

Urban environments offer a challenging scenario for autonomous driving [6]. The proposed solution allows autonomously navigate urban roadways with minimum a priori map or GPS. Localization is achieved by Kalman filter extended with odometry, compass, and sparse landmark measurement updates. Navigation is accomplished by a compass-based navigation control law. Experiments validate simulated results and demonstrate that, for given conditions, an expected range can be found for a given success rate.

The architecture contains steering and speed controllers, an object tracker, a path generator, a pose estimator, and a navigation algorithm using sensors allowing real-time control. High-level localization is provided by the pose estimator, which utilizes only odometry measurements, compass measurements, and sparse map-based measurements. The sparse map-based measurements generated from computer vision methods compare raw camera images to landmark images contained within a sparse map. The roadway scene includes lane line markings, road signs, traffic lights, and other sensor measurements. The scene information and the inertial pose estimate are fed into a navigation algorithm to determine the best route required to reach the target. This navigation scheme is provided by a compass-based navigation control law.
