Managing automatic vehicles in road networks is a major challenge with significant scientific and practical relevance. The mixed traffic scenario, which has both automated vehicle and conventional vehicles in road network, is one issue that necessitates special attention during the introductory phase of automated vehicles. In this paper we propose a strategy for managing automated vehicles in a mixed traffic scenario. By the proposed strategy, automated vehicles will form platoons, in which the first automated vehicles are manually driven. Specifically, one target automated vehicle can decide to join one proceeding platoon or to lead other automated vehicles after entering the managed road network. A deep-Q-learning method is adopted to find optimal decisions of automated vehicles over space and time within their trips. In addition, the optimization with three different objectives will be investigated, which are achieving the longest average automated driving duration of all automated vehicles, achieving the lowest average travel delay of all automated vehicles and achieving the lowest overall travel delay. A 3x3 grid network with 9 signalized intersection is examined in simulation. The results show that with a longest automated driving duration achieved, a longest overall travel delay will be conducted; with a shortest individual travel delay achieved, a shortest automated driving duration will be conducted. However, with a shortest overall travel delay achieved, automated vehicles compromise few for its own benefit (with a minor deterioration in automated driving duration and travel delay compare to other objective optimizations). We further conclude that coordinating automated vehicles to form platoons to achieve a shortest overall delay can be beneficial for both individual automated vehicle drivers and the overall traffic.

CICTP