OnSite平台自动泊车工具建设与思考

Simulation testing, recognized for its cost-efficiency and effectiveness, has emerged as an indispensable methodology for addressing the challenges associated with performance verification in autonomous driving. In an effort to further augment the capabilities of the OnSite simulation platform, Professor Zheng Sifas team at Tsinghua University has devised and introduced an automated parking evaluation tool specifically designed for autonomous valet parking technologies. Initially, the team proposed a method for constructing a scenario library, which is based on the collection and modeling of real vehicle data. This approach integrates the latest industry standards with collected data on parking spaces to establish a more comprehensive set of test scenarios. Subsequently, a multidimensional parking evaluation method was designed, emphasizing completion rates while concurrently considering safety, efficiency, and accuracy. Moreover, the proposed evaluation tool was subjected to Hardware-in-the-Loop simulation testing and was benchmarked against results from the mainstream CARLA simulation platform as well as actual vehicle testing platforms to ascertain its effectiveness. The study also examines the outcomes of the second OnSite Autonomous Driving Algorithm Challenge, particularly focusing on the autonomous parking. This examination involved analyzing the scoring and deductions of the top ten teams to elucidate the strengths and weaknesses of current autonomous parking algorithms in practical applications. Lastly, by synthesizing competition outcomes with prevailing technological trends, this study provides a summary of pertinent experiences and offers insights and future prospects for the development of autonomous driving parking evaluation tools and the enhancement of the OnSite platform. The research findings hold considerable significance for the advancement of autonomous parking technology and the progression of autonomous driving simulation testing.