Publication

Autonomous driving cars have been a hot topic in the media in recent years, with more and more tech companies and universities presenting projects with fully automated vehicles. Most of these vehicles rely on highly sophisticated and expensive sensors that are not yet feasible for commercial vehicles. On the other hand automated systems that are implemented in commercially available vehicles are largely limited to active safety scenarios where the system only assists the driver in dangerous situations such as collision avoidance or lane support. The goal of this thesis project is to use sensors on commercially available vehicles for automating lateral control in selected scenarios. The evaluated scenarios are limited to roads where the sensors can detect lane markings or a preceding vehicle or both. The approach was to generate two different reference paths, one from lane markings and one from the preceding vehicle information. The lane marking path is generated from filtering measurements from the lane detection cameras using a nonlinear Kalman filter. The preceding vehicle path is generated by fusing the radar and camera measurements of the preceding vehicle. In order to develop and tune the filters a detailed analysis was done on the sensor measurements collected for this project. The implemented filters improve the current system in several ways. When the lane marking are lost for short period of time the prediction from the last measurement update can provide a reference while driving up to 40 meters. The path generated from the estimates of the preceding vehicle describe the trajectory which the vehicle has driven. This way a more accurate reference signal can be generated than using only the current position of the preceding vehicle, especially in turns and at long distances. By having two references paths the lateral control is more robust and an algorithm that takes the covariances of the estimation for both paths into account guarantees a smooth transition between them.