Reinforcement Learning from Scarce Experience via Policy Search L