Path Planning Model Based on Constrained Policy Iteration

With the increasing severity of traffic congestion, the frequency of traffic emergencies has also increased. When an emergency occurs, it is necessary to send a large number of emergency vehicles from all over the city to the rescue point, therefore, it is necessary to plan the best feasible path from multiple locations to the same destination quickly and efficiently. At the same time, emergencies have a serious impact on urban operation. The emergency management of urban traffic puts forward more and more specific conditions and requirements around path planning: for example, find the path with as few intersections as possible, and find multiple paths with repetition rate constraints. Because the existing intelligent algorithms such as Dijkstra algorithm and PSO algorithm can only find one optimal path from the starting point to the target point, and the unconstrained reinforcement learning algorithm can find multiple paths with poor performance and can not meet the specific traffic display requirements, this paper proposes a model based on constrained policy optimization to solve the path optimization problem in the context of traffic emergency. The model is a continuation of the policy iteration algorithm. When the policy is improved, the action is constrained, and the policy with the largest cumulative reward under the constraints is selected. It can solve multiple paths from the starting point to the target point, and find the path that meets the step constraint and the repetition rate constraint with the known path. When solving the problem of finding the path from multiple points to the target point, the time efficiency is higher than Dijkstra algorithm.