Decision Frameworks for Using Uncertain Predictions for Cut-In Detections in (Semi-) Automated Driving

This work examines how a classifier's output of a cut-in prediction can be mapped to a (semi-) automated car's reaction to it. Several approaches of decision making are compared using real world data of a lane change predictor for an automated longitudinal guidance system, similar to an adaptive cruise control system, as an example. We show how the decision algorithms affect the time when a new lead vehicle is selected and how much more comfortable we can decelerate given different selection strategies. We propose a novel decision algorithm and conducted a case study with a prototype research car to evaluate the subjective quality of the different approaches.