Robust predictive cruise control for commercial vehicles

In this paper we explore learning based predictive cruise control and the impact of this technology on increasing fuel efficiency for commercial trucks. Traditional cruise control is wasteful when maintaining a constant velocity over rolling hills. Predictive cruise control is able to look ahead at future road conditions and solve for a cost effective course of action. Model based controllers have been implemented in this field but cannot accommodate the many complexities of a dynamic environment which includes changing road and vehicle conditions. In this work, we focus on incorporating a learner into an already successful model based predictive cruise controller in order to improve its performance. We explore back propagating neural networks to predict future errors then take actions to prevent said errors from occurring. The results show that this approach improves the model based predictive cruise control by up to 60% under certain conditions. In addition, we explore the benefits of classifier ensembles to further improve the gains due to intelligent cruise control.