Adaptive System for Autonomous Driving

Avoiding faults in systems is of uttermost importance during system development. In case of autonomous systems this requirement becomes more important because of the fact that there is no human in the loop that can take over control after a fault. In this paper, we discuss a methodology allowing to implement a system that reacts on faults in a smart way using rules specifying possible sequences of actions a system can take for reaching a goal. In the methodology the selection of actions is done automatically aiming at reaching the goal state. The methodology allows a system to adapt in cases where action execution fails. Besides the underlying foundations, we show the applicability of the methodology using an example from the autonomous driving domain considering different sensors for obstacle detection. For this case study the methodology leads to a substantial improvement of availability compared to a random selection approach.