Event-Driven Stochastic Eco-Driving Strategy at Signalized Intersections From Self-Driving Data

Fuel consumption and travel time of a vehicle are significantly influenced by driving behavior, especially when approaching a signalized intersection. Injudicious driving reacting to sudden changes in traffic signal can lead to additional energy consumption and increase of travel time. This paper presents a learning-based event-driven ecological (eco) driving system (EDS) that generates the optimal velocity from self-driving data of a vehicle. Currently, full autonomy of vehicles and proper infrastructure development for vehicle-to-vehicle and infrastructure-to-vehicle communications are not widespread; however, the proposed system can be beneficial for driving scenarios in the existing traffic environment. We design a Gaussian process model using a Bayesian network for naturalistic learning from driving data and traffic signal condition to estimate the probability of a vehicle crossing the intersection within a signal phase. Based on the estimated probability, the optimal velocity is generated and the vehicle (driver) will be advised to either slow down earlier (to avoid aggressive braking) at the red signal or speed up (to cross the intersection) at the green signal. Finally, microscopic simulations are performed to evaluate the performance of the proposed scheme. The results show significant performance improvement in both fuel economy and travel time.