Minimizing Response Times in Real Time Planning and Search

Real time artificial intelligence (AI) systems are required to respond within a given deadline, or have optimal response times. While some researchers have addressed the issue of planning under deadline constraints, there has been very little research towards optimizing the response time of problem-solving methods. The costs for a response consists of the cost to plan for a solution and the cost of executing the chosen solution. There is an intimate trade-off between these two costs. This paper presents an algorithm for providing near optimal response times by formalizing the trade-offs between planning and execution costs. We provide a proof of correctness and describe an implementation of the algorithm in a real time application of query planning. We also provide a model for considering response times in the context of the A* heuristic search algorithm.