Approximate dynamic programming for the solution of multiplatform path planning problems

Abstract

We consider the problem of planning the paths of multiple vehicles in observing a battle space with the possibility of vehicle destruction. We illustrate the significant complexities that arise when stochastic effects (i.e. random vehicle destruction) are introduced into the model. Dynamic programming is the classical framework that characterizes solutions to our problem and drives the algorithmic development. While computationally expensive, the dynamic programming recursion can be employed to solve the stochastic problem directly. Similarly, dynamic programming can be used to solve various deterministic auxiliary problems whose solutions provide heuristic solutions to the original stochastic problem. We describe our approach and illustrate preliminary results.