h1

%0 Thesis
%A Scheffe, Patrick
%T Prioritized motion planning for connected vehicles
%V 2025,01
%I RWTH Aachen University
%V Dissertation
%C Aachen
%M RWTH-2025-07834
%B Aachener Informatik-Berichte
%P 1 Online-Ressource : Illustrationen
%D 2025
%Z Veröffentlicht auf dem Publikationsserver der RWTH Aachen University
%Z Dissertation, RWTH Aachen University, 2025
%X Networked control is concerned with the control of multiple agents which interact. When formulating the control problem as an optimization problem, the highest solution quality can be achieved by centralized control which considers all agents at once. However, the number of problem dimensions then grows with the number of agents. When solving nonconvex optimization problems, the computation time scales non-polynomially with the number of problem dimensions. Combining centralized multi-agent control with nonconvex optimization, a time-constrained optimization is often intractable. A relevant application is multi-agent motion planning (MAMP) for connected and automated vehicles (CAVs) with collision avoidance. We address the nonconvexity of the optimization problem by modelling the system dynamics using an automaton, enabling the use of graph-search algorithms for motion planning. Our motion planning algorithm has a limited horizon to decrease computational complexity and obtains a constant computation time with a fixed number of search operations. We achieve recursive feasibility through the structure of the automaton modelling the system dynamics. We address the growing number of problem dimensions in a centralized formulation with prioritized planning (PP). In PP, agents plan sequentially according to their priorities, and lower-priority agents respect the plans of higher-priority agents. This work addresses three challenges in PP. First, the effect of the prioritization on the solution quality and the computation time is often unknown a priori. We present three prioritization approaches: one using a heuristic to increase solution quality, one to minimize the number of sequentially planning agents and thus the computation time, and one to simultaneously compute multiple prioritizations, which can increase solution quality or reduce computation time. Second, the computation time in PP grows approximately linearly with the number of agents, which poses a challenge in large-scale systems. We present an approach to bound the computation time as desired, while improving the solution quality compared to an approach in which all agents compute in parallel. Third, an agent can fail to find a feasible solution in PP. We present an approach to ensure recursive feasibility for every agent. When testing algorithms for networked control, it is difficult to achieve reproducibility of the experiments. We present a software architecture to address this challenge and implement it in our Cyber-Physical Mobility Lab (CPM Lab). The CPM Lab is an open-source testbed for CAVs. We evaluate our methods on prioritization and bounded computation time for prioritized motion planning in numerical and physical experiments in the CPM Lab.
%F PUB:(DE-HGF)11 ; PUB:(DE-HGF)3
%9 Dissertation / PhD ThesisBook
%R 10.18154/RWTH-2025-07834
%U https://publications.rwth-aachen.de/record/1018409