9. Digital Maps for Driving Assistance Systems and Autonomous Driving

Modern passenger vehicles increasingly incorporate perception systems that allow the deployment of advanced driving assistance systems which in turn shall led to highly automated systems and ultimately to autonomous vehicles. Despite the numerous advances in sensors and communications technologies applied to passenger vehicles, perception remains a major challenge due to the complexity of the task, the vehicle geometric constraints and cost. Digital navigation maps have proven themselves to be essential for driver guidance and have replaced paper maps. They store the geometric description of roads and associated features. Due to the limits of perception systems, errors occur on the understanding and relevance of the detected objects, when using multiple sensors discordances occur that lead to unknowns. By projecting sensor information from the perceived environment into digital navigation maps, information can be contextualized. The resulting representation of the world is much easier to interpret by a machine, but also to ensure the coherence of the perceived information.

This article deals with two aspects of perception: situation understanding and the detection of errors in maps. One goal is to demonstrate how contextualized information can be used to facilitate situation understanding, that is to provide meaning to the spatio-temporal relationship between perceived objects, road features, and the subject vehicle. Situation understanding is a fundamental feature for a machine to take decisions in an appropriate manner, particularly, if the navigation of the vehicle is to be governed by it. The approach is based on the application of ontologies to facilitate contextual descriptions by enabling the formalization of information in a semantic manner. The digital map on board the vehicle stores a priori knowledge about the environment, that is the underlying structure and static context. Digital maps contain different errors, particularly regarding road geometry. In case of errors, the association between the perceived objects and maps for situation understanding will be wrong. This will propagate through the system and lead to hazardous situations. The chapter is thus completed by introducing a mathematical formalism that allows for the detection of geometric map errors in real time. The uniqueness of this formalism is that it uses production type components, namely GNSS receivers and vehicle state data. The approach includes a mechanism that allows for the identification of the error, which could be in the digital map itself or due to errors in the position estimates of the vehicle.

The chapter combines theoretical developments with experimental data, as the proposed solutions were tested in public road networks using purposely equipped vehicles to demonstrate the viability and advantages of the theoretical developments.