Optimizing Fixed-Size Stochastic Controllers for POMDPs

In this paper, we discuss a new approach that represents POMDP policies as finite-state controllers and formulates the optimal policy of a desired size as a nonlinear program (NLP). This new representation allows a wide range of powerful nonlinear programming algorithms to be used to solve POMDPs. Although solving the NLP optimally is often intractable, the results we obtain using an off-theshelf optimization method are competitive with state-of-theart POMDP algorithms. Our approach is simple to implement and it opens up promising research directions for solving POMDPs using nonlinear programming methods.