Reconciling strategic and tactical decision making in agent-oriented simulation of vehicles in urban traffic

We consider an integrated decision making process of autonomous vehicles in agent-oriented simulation of urban traffic systems. In our approach, the planning process for a vehicle agent is separated into two stages: strategic planning and tactical planning. During the strategic planning stage the vehicle agents constructs the optimal route from source to destination; during the tactical planning stage the operative decisions such as speed regulation and lane change are considered. For strategic planning we modify the stochastic shortest path algorithm with imperfect knowledge about network conditions. For tactical planning we apply distributed multiagent reinforcement learning with other vehicles at the same edge. We present planning algorithms for both stages and demonstrate interconnections between them; an example illustrates how the proposed approach may reduce travel time of vehicle agents in urban traffic.