POMDPs

What is the problem being addressed? The paper addresses the problem of solving large partially observable Markov decision processes (POMDPs) in an anytime manner. Why is that a problem worth solving?