von Herrn Muhammet Öncel
Kurzfassung/Abstract:
An autonomous vehicle follows the trajectory that its motion planner provides, but a planned trajectory can still be unsafe. This thesis presents a module for the evaluation of planned trajectories during runtime, independently of the planner. The evaluation uses four criticality metrics from the literature, which measure how critical a traffic situation is. Each of them is designed for certain types of situations, such as following a vehicle ahead: Time-To-Collision (TTC), Time Headway (THW), Deceleration Rate to Avoid a Crash (DRAC), and the safe distances of Responsibility-Sensitive Safety (RSS). Each metric's result is expressed in Subjective Logic, a framework that states not only how safe a trajectory is, but also how certain the evaluation is. When a metric's assumptions do not hold for a situation, its result is uncertain instead of safe, so a lack of information is not confused with the absence of danger. The results of all metrics are then combined into one evaluation of the situation.
In the talk, I will explain how each metric's result is expressed in Subjective Logic and how the results are combined. I will then present the evaluation in simulated traffic situations: how well the combined result detects danger, and which metrics work best in which situations.
