A parallel algorithm for POMDP solution

Abstract

Most exact algorithms for solving partially observable Mar-kov decision processes (POMDPs) are based on a form of dynamic pro-gramming in which a piecewise-linear and convex representation of the value function is updated at every iteration to more accurately appro-ximate the true value function. However, the process is computationally expensive, thus limiting the practical application of POMDPs in plan-ning. To address this current limitation, we present a parallel distribu-ted algorithm based on the Restricted Region method proposed by Cas-sandra, Littman and Zhang [1]. We compare performance of the parallel algorithm against a serial implementation Restricted Region.