Learning Generalized Reactive Policies using Deep Neural Networks