Collective Robot Reinforcement Learning with Distributed Asynchronous Guided Policy Search