Proposing Cooperative Maneuvers Among Automated Vehicles Using Machine Learning
Bernhard Häfner, Josef Jiru, Henning F. Schepker, Georg Schmitt, Jörg Ott
Abstract
Cooperative maneuvers will enable automated vehicles to optimize traffic flow and increase safety via vehicle-to-vehicle communication. Different approaches and protocols exist, but no study has investigated how to generate intelligent suggestions for cooperative maneuvers. We use machine learning to propose safe and suitable overtake maneuvers. To this end, we train a classifier for maneuver success as well as regression models on an extensive data set of randomized initial situations. In addition, we show that changing objective functions allows optimizing for different goals like smoothness or driven distance. Our evaluation shows that machine learning is well-suited to suggest cooperative maneuvers while also facing some trade-offs. This work may thus provide a benchmark for advanced studies on cooperative maneuver proposals.