On the Scalability of Parking Trajectory Optimization of Autonomous Ground Vehicles

Although the use of optimization-based algorithms for autonomous motion planning in the context of parking has been studied in the literature, most of the existing works were based on either unrealistic simulation environments or a single vehicle type or model. In order to support the deployment of such frameworks for real-world applications, the need for the scalability analysis of such optimization frameworks under realistic simulation environments as well as different vehicle types becomes important. Therefore this paper investigates the suitability of a two-stage optimization framework under a realistic simulation environment as well as using 4 different vehicle models. Specifically, the two-stage optimization process involves first the use of the A star algorithm for initial path generation, and in the second stage, Sequential Quadratic Programming (SQP) is used to optimize the results pathways. In terms of vehicle type, we employ four different vehicle types with different model parameters and evaluated the performance of the framework accordingly. The results show that also the optimization framework is capable of generating feasible parking trajectories, some vehicle types require more script run-time compared to others.