Towards Direct Policy Search Reinforcement Learning for Robot Control