Energy-Efficient Motion Planning for Autonomous Vehicles Using Uppaal Stratego

Energy-efficient motion planning for autonomous battery-powered vehicles is crucial to increase safety and efficiency by avoiding frequent battery recharge. This paper proposes algorithms for synthesizing energy- and time-efficient motion plans for battery-powered autonomous vehicles. We use stochastic hybrid games to model an appropriate abstraction of the autonomous vehicle and the environment. Based on the model, we synthesize energy- and time-efficient motion plans using Q-learning in Uppaal  Stratego. Via experiments, we show that pure Q-learning is insufficient when the problem becomes complex, e.g., Motion Planning (MOP) in large environments. To address this issue, we propose Concatenated Motion Planning (CoMOP), which divides the environment into several regions, synthesizes a motion plan in each region and concatenates the local plans into an entire motion plan for the whole environment. CoMOP enhances the applicability of Q-learning to large and complex environments, reduces synthesis time, and provides efficient navigation and precise motion plans. We conduct experiments with our approaches in an industrial use case. The results show that CoMOP outperforms MOP regarding synthesis time and the ability to deal with complex models. Moreover, we compare the energy- and time-efficient strategies and visualize their differences on different terrains.