Publication

The SALI project aimed at investigating how challenging runtime factors such as unpredictability, faults and limited resources affect and could be managed in self-driving vehicles (SDVs). In order to achieve that, a series of experiments have been conducted on the AstaZero test track where three main use cases were tested: a road accident, a sensor fault and critical battery level. Experiments were conducted in the city area as well as on the rural road. Three vehicles were utilized during the experiments; one of them running the SALI software solution. In the three use cases, the SALI vehicle had the objective of supporting the driver on his/her daily journey from work (i.e., an origin point) to home (i.e., a destination point). Machine learning techniques were applied to find patterns on driver’s behavior (e.g., preferred routes). Different scenarios have been tested for each use case, for instance, with and without traffic. Thanks to the learning-based adaptive behavior enabled by our software solution, the SALI vehicle was able to timely react to the different runtime events, ensuring the self-driving functionality availability, adequacy and resilience. The results demonstrate the importance of correctly managing runtime factors that could affect SDVs, and the validity of our solution for correctly supporting this task.