A distributed asynchronous algorithm for expected average cost dynamic programming

A distributed asynchronous implementation of the value-iteration algorithm in dynamic programming is presented. The iteration step is carried out by a number of processors, each iterating on a subset of the value function vector. Each processor transmits its computed coordinates to other processors. The algorithm converges when the different processors iterate at different speeds. The information received by a processor regarding other coordinates may be outdated, and there may be an unpredictable delay in receiving information from other processors.<
>