Around View Monitoring System based Vehicle Localization and Model Predictive Control for Automated Vehicle in Urban Roads

Abstract

Recently, major corporations have announced plans to begin selling automated vehicles in a few years, and some jurisdictions have passed legislation to allow such vehicles to operate legally on public roads. Even, some leading companies release new vehicles providing an opportunity to experienc¬e partially automated driving. Distronic Plus (Mercedes-Benz), Driving Assistant Plus (BMW), Highway Driving Assist (Hyundai Motor Company) are typical examples of partially automated driving system. With these system, a car can maintain a distance from a vehicle in front of it without the driver doing anything. It only works when the driver holds the steering wheel. When the driver takes their hands off for a certain time, the function is disabled. In spite of operating limit, they must be offering increased comfort and safety. The main reason of limited operation is that the environment sensing system available for production cars still do not reach a satisfying level of development in terms of robustness and availability at various road conditions. Actually, many of these problem could be solved by using the current state-of-the-art sensor such as highly accurate inertial navigation systems and 3D scanning laser rangefinders However, integrating them into cars will increase the price and represent yet another barrier to adoption. Therefore, this dissertation focused on developing a fully automated driving algorithm which is capable of automated driving in urban roads while a chosen sensor configuration is closer to current automotive serial production in terms of cost and technical maturity than in many autonomous vehicles presented earlier. Mainly two research issues are considered: a lane-level localization and a model predictive vehicle control. The lane-level localization implies positioning the vehicle with centimeter-level accuracy with respect to a map. In order to achieve a satisfactory level of position accuracy with a low-cost GPS, a sensor fusion approach is essential for lane-level localization. The proposed sensor fusion approach for the lane-level localization of a vehicle uses an Around View Monitoring (AVM) module and vehicle sensors. The proposed algorithm consists of three parts: lane detection, position correction, and localization filter. In order to detect lanes, a commercialized AVM module is used. Since this module can acquire an image around the vehicle, it is possible to obtain accurate position information of the lanes. With this information, vehicle position can be corrected by the iterative closest point (ICP) algorithm. This algorithm estimates the rigid transformation between the lane map and lanes obtained by AVM in real-time. The vehicle position corrected by this transformation is fused with the information of vehicle sensors based on an extended Kalman filter (EKF). For higher accuracy, the covariance of the ICP is estimated using Haralicks method. In contrast with highway, automated vehicles are allowed a relatively short headway distance in urban roads. Accordingly, the information from the environment sensor such as radars, lidars and cameras was very limited due to neighboring vehicles. In such situations, suddenly appeared obstacles may put automated vehicles in danger. To overcome this problem, preceding vehicles behavior information was used for automated vehicle control in this dissertation. Preceding vehicle Behaviors including a yaw motion was precisely estimated using front lidars. It makes possible to generate the trajectory for following the preceding vehicle. An optimal vehicle following problem while avoiding collision with obstacles is formulated in terms of cost minimization under constraints. To solve this minimization problem, we use model predictive control approach. The performances of the proposed localization and control algorithm of automated vehicle are verified via computer simulations and vehicle tests. Test results show that the proposed methods can achieve centimeter-level localization accuracy and robustness of automated vehicle control system in urban roads.