POMDPs

Point-based value iteration: An anytime algorithm for 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?