Assessment and optimization of automation processes in environments used for Autonomous Driving

Abstract

Testing autonomous driving functionalities is a very important but complicated task. Due to the high costs of real-life testing, testing in virtual machines has become a must in the industry. At Volvo Cars, a virtual testing toolchain where traffic scenarios are generated, simulated, and analyzed is utilized. However, due to a large number of test cases, this is a very costly process. This thesis aims to optimize the execution time of the testing toolchain to cut down costs. The study started with profiling the toolchain to identify bottlenecks and areas where execution time could be shortened. Based on the profiling, a concept of a ”monitor” was proposed, which analyzes scenarios in real-time rather than sequentially. The study implemented a proof of concept of a ”stay-in-lane monitor” which checks whether the monitor remains within its original lane at every time step of the simulation. The results showed that the ”stay-in-lane monitor” significantly reduced the toolchain’s execution time. Based on these results, the ”monitor” concept was deemed successful, and the team at Volvo Cars will continue to implement more monitors in the future. We conclude that the optimization approach presented will help the collaborating team achieve more efficient testing, thus reducing costs and improving the development of autonomous drive functionalities.