Platooning Control with Deep Reinforcement Learning

Abstract

Considering an increasing urbanization on a global scale, traffic congestion is becoming a progressively larger problem. At the same time, new communication standards enable an interconnectedness of vehicles, which can be used to realize highly automated driving. Highly automated driving allows little distances to the preceding vehicle, thereby increasing the road capacity and decreasing the energy consumption. Artificial intelligence made great advances in recent time and can now be used for ambitious control problems. In this work, the usability of deep Reinforcement Learning for platooning control is examined and the possibility of energy minimization whilst still ensuring operational safety is investigated. Alongside string stability, also robustness against burst errors in communication is incorporated. The obtained controllers are verified with respect to given specifications and validated by comparison with a model-based controller.