Autonomous vehicles will share public roads with human-driven vehicles for the foreseeable future. Motion planning must therefore always take into account the behavior of human drivers in the environment. The estimation of the current and the prediction of the future driver behavior is achieved using driver behavior models. They usually represent general driver behavior, neglecting location-dependent aspects. Following the state of the art, they are learned based on measurements of real-world traffic. Today's production vehicles are often already able to continuously measure their own position due to their equipment with GNSS and environmental sensors. The simultaneous increase of networking technology in vehicles offers the potential to use them as a measurement fleet to record extensive data sets of real-world traffic. For driver behavior modeling, this wide availability of motion data opens up the possibility of modeling driver behavior in great detail, even covering location-dependent aspects. This dissertation develops a driver behavior model that optionally makes use of prior knowledge about location-dependent driver behavior. It consists of two model parts: The first part, the situation model, models the behavior of a human-driven vehicle in a general form using the route and the trajectory. The model is taken from the state of the art and adapted. As input features of the predicted vehicle, it uses its motion state, the geometry of its route, and its relationship to other vehicles in the environment. Provided a digital map of the road network, the model is location-independent. The second part, the local maneuver model, represents the local maneuver probability at individual junctions in the road network. It is integrated into the state estimation of the situation model depending on availability. Both the adopted approach of the local maneuver model and the integration of location-dependent prior knowledge into a general driver behavior model represent a technical novelty. For analysis and modeling of driver behavior, the inD and rounD datasets of vehicle trajectories are used in this work. Both are taken from recordings of real-world traffic at selected intersections and roundabouts. From an analysis of driver behavior at junctions, the distribution of maneuvers is found to be strongly dependent on the location. In the modeling of driver behavior, this is taken into account by extracting a local maneuver model. An evaluation shows, that the addition of the local maneuver model to the situation model proves to be advantageous, especially at junctions with a strongly uneven maneuver distribution.
Modellierung des ortsabhängigen Fahrerverhaltens und Integration in ein allgemeines Modell zur Bewegungsprädiktion von Kraftfahrzeugen
Modeling location-dependent driver behavior and integration into a general model for motion prediction of motor vehicles
01.01.2023
1 Online-Ressource : Illustrationen pages
Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2023; Aachen : RWTH Aachen University 1 Online-Ressource : Illustrationen (2023). = Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2023
Sonstige
Elektronische Ressource
Deutsch
Modellierung des taktischen Fahrerverhaltens bei Spurwechselvorgängen
Tema Archiv | 2003
|Modellierung des Fahrerverhaltens: Regel- und Steuerstrategien
Tema Archiv | 1992
|