Evaluation of two Cooperative Maneuver Planning Approaches at a Real-World T-Junction in Mixed Traffic

Connected automated driving promises a significant improvement of traffic efficiency and safety on highways and in urban areas. Cooperative maneuver planning at unsignalized intersections may facilitate active guidance of connected automated vehicles. Previous such works mostly employ simple rule-based or optimization-based approaches, often only for fully automated vehicles and only in simulated environments. In this article, we extend and evaluate our previously introduced approaches, which -- in contrast -- are capable of handling mixed traffic, i.e., automated vehicles and regular vehicles driven by humans sharing the road. They are based on a multi-scenario prediction and on graph-based reinforcement learning, respectively. For the first time in literature, we thoroughly evaluate cooperative planners in a high-fidelity simulation with fully automated traffic and mixed traffic using state-of-the-art human driver models and real-world automation software. In addition, we are the first to present respective real-world evaluations with three prototype automated vehicles in public traffic, which confirm the simulative results. Our quantitative evaluations show that cooperative maneuver planning achieves a significant reduction of crossing times and the number of stops even in a realistic environment with only few automated vehicles.