This report introduces a new approach to solve sensor management problems. Classically sensor management problems are formalized as Partially-Observed Markov Decision Process (POMPD). Our original approach consists in deriving the optimal parameterized policy based on stochastic gradient estimation. Two differents techniques nammed Infinitesimal Approximation (IPA) and Likelihood Ratio (LR) can be used to adress such a problem. This report discusses how these methods can be used for gradient estimation in the context of sensor management . The effectiveness of this general framework is illustrated by the managing of an Active Electronically Scanned Array Radar (AESA Radar).