Model-based certification of automated vehicles

Abstract

Automated vehicles (AVs) have long been predicted to disrupt the transportation industry any moment. Although numerous companies have shared that optimism and supported development, it now seems that the challenges of building automated vehicles are becoming apparent and are pushing the vision far into the future. While manufacturers are making technological progress there are concerns about the safety of AVs. For AVs new types of accidents such as accidents due to unsafe software behavior and interactions between vehicle hardware, software, and humans must be considered in addition to what has been done for the safety of conventional vehicles. Many stakeholders in the AV industry are looking for ways to ensure and demonstrate the safety of AV designs. For one, regulators are responsible for assessing safety and granting certification based on whether a minimum level of confidence in safety was achieved.

Similarly, manufacturers need to be able to argue for the safety of their AV design in order to receive certification and convince customers. Finally, suppliers such as software/hardware providers, AV app developers, etc., also depend on a way to demonstrate the safety of their product and convince Original Equipment Manufacturers (OEMs). Hence, safety is an important problem for the progress of the AV industry that calls for a solution. This thesis presents a potential solution by suggesting a model-based certification. First, currently existing problems with AV safety and the currently taken approaches to address AV safety are identified. Then, an alternative model-based safety approach that addresses the identified problems is demonstrated and a conceptual architecture model and safety requirements are derived. Finally, it is shown how the model-based safety approach may be implemented through certification.

The application to safety-related problems such as regulation and the insurance of AVs is outlined and an organizational structure and processes for model-based certification are derived.