@phdthesis{,
    author = {Heinrich, Benjamin Christian},
    title = {Towards Smoother and More Precise Autonomous Driving},
    editor = {},
    booktitle = {},
    series = {},
    journal = {},
    address = {},
    publisher = {},
    edition = {},
    year = {2023},
    isbn = {},
    volume = {},
    number =  {},
    pages = {},
    url = {},
    doi = {},
    keywords = {Motion Planning, Motion Control, System Architecture, Autonomous Driving},
    abstract = {Autonomous driving evolved into a vast field of research. While it started already in the 80s of the last century, it was really kicked off with the DARPA Challenges (Grand and Urban, respectively). Now, fully automated Highway Pilots and Urban Automated Taxis seem within reach as different major players in the industry announce these regularly. Yet, as regularly, these plans are postponed which shows that there are still more than enough problems to solve. In this work, planning and control problems for autonomous driving in unstructured environments, e.g., gravel roads through the woods or very dense urban scenarios, are considered. The goal is to achieve improved driving performance in terms of smoothness and precision especially in these challenging settings. Since every part of the control chain - from perception over fusion and planning to actuation - can only be as strong as its weakest link, this work starts with the system architecture. A new architecture is proposed which, by providing the latter parts of the control chain with more information, enables better performance of the planning and control part and thus improves the overall performance. The new architecture also aims at facilitating development and maintenance. Simple practical examples show improved driving performance, even with the same controllers in place. One main focus in this work is motion planning. The backbone of the TAS (Institute for Autonomous Systems Technology) motion-planning framework is an extension of the so-called hybrid-state A* algorithm where a guided exploration is used. Further, the classic path-velocity decomposition is used and extended by using it in conjunction with modern trajectory-planning and optimization techniques. For free driving, i.e., without lead vehicle, state-of-the-art collision-checking methods are improved. For convoy driving, i.e., one vehicle following another, TAS’s own award-winning following performance is improved upon. For platooning, i.e., multi-vehicle convoys, the state of the art is defined in a novel setting: without any lateral guidance from lane markings and with only a shared bandwidth of 19.2 kbit/s for all inter-vehicle communication. The other main focus in this work is motion control. The new architecture has a vehicle-specific low-level part and a vehicle-unspecific high-level part. While the former is described mainly for completeness, the latter features several contributions. The main mode of operation is trajectory following, where the importance of localizing the ego vehicle in both time and place consistently over time is stressed. This allows, together with the consistency of the trajectory generation introduced in the former chapter, to handle delays in the control chain efficiently. Further, extensions to the trajectory following - especially for the convoying use-case - are discussed. Finally, it is described how a normal-sized (electric) vehicle is guided and positioned with sub-centimeter precision in outdoor scenarios.},
    note = {},
    
    school = {Universität der Bundeswehr München},
    
}