Automated Monitoring of Traffic Sequence Chart Specifications for Simulated Driving Scenarios

Abstract

This thesis addresses the verification of the reliability of autonomous driving cars. More specifically, automated verification is examined for satisfaction of simulated driving scenarios. For this purpose, a specific functional scenario (stopping of a car within a certain distance in front of an obstacle) is formalized into a Traffic Sequence Chart. In addition, concrete scenarios with expected satisfaction values for the functional scenario are developed based on the functional scenario using requirements engineering techniques. The Traffic Sequence Chart is validated by running several simulations with these concrete scenarios in the CARLA simulation, and evaluating the Traffic Sequence Chart with the simulated sensor data for satisfaction. A Traffic Sequence Chart runtime monitor is developed, and it's correctness partially shown, that processes the sensor data from the CARLA simulation and checks automatically whether the sensor data satisfies the Traffic Sequence Chart. The input sensor data for the monitor is received via ROS from the CARLA simulation using the ROS bridge, and the execution of the monitor takes place in a ROS node. Due to optional noisy sensor data from the simulation and the sampling rate of the simulation, the monitor has adjustable tolerances. The evaluation of the Traffic Sequence monitor has shown that simulated concrete scenarios can be classified according to the expected satisfaction value. The time between the verdict of the monitor and the theoretical minimum time with the sensor data is less than 0.4 seconds. The limit of the monitor was examined by designing a concrete scenario with consideration of the tolerances, which is incorrectly classified as satisfied by the monitor.