Comparison of Architectural Design Decisions for Resource-Constrained Self-Driving Cars - A Multiple Case-Study

Context: Self-Driving cars are getting more and more attention with public demonstration from all important automotive OEMs but also from companies, which do not have a long history in the automotive industry. Fostered by large international competitions in the last decade, several automotive OEMs have already announced to bring this technology to the market around 2020. Objective: International competitions like the 2007 DARPA Urban Challenge did not focus on efficient usage of resources to realize the self-driving vehicular functionality. Since the automotive industry is very cost-sensitive, realizing reliable and robust selfdriving functionality is challenging when expensive and sophisticated sensors mounted very visibly on the vehicle’s roof for example cannot be used. Therefore, the goal for this study is to investigate how architectural design decisions of recent self-driving vehicular technology consider resource-efficiency. Method: In a multiple case study, the architectural design decisions derived for resourceconstrained self-driving miniature cars for the international competition CaroloCup are compared with architectural designs from recent real-scale self-driving cars. Results: Scaling down available resources for realizing self-driving vehicular technology puts additional constraints on the architectural design; especially reusability of software components in platform-independent algorithmic concepts are prevailing. Conclusion: Software frameworks like the robotic operating system (ROS) enable fast prototypical solutions; however, architectural support for resource-constrained devices is limited. Here, architectural design drivers as realized in AUTOSAR are more suitable.