Debugging Autonomous Driving Systems Using Serialized Software Components

In the development of software-intensive systems in a vehicle, like an autonomous driving system, defects are often only recognized during trials on the physical vehicle. In contrast to a simulation environment, a physically executed maneuver does not offer the possibility to pause and debug critical code sections or to reproduce and repeat faulty trials. Furthermore, development space and capacities are limited inside the car. Therefore, it is best practice to analyze faults observed during a physical execution offline and to reproduce faulty trials in a simulation environment. The repetition in a simulation environment is a time consuming effort but necessary for pushing the software component towards a state in which it showed the faulty behavior. This paper shows an approach for executing the faulty state again in a simulation environment by serializing the exact state of the software system and summarizes practical experience gained by this approach.