An Actor-Critic-Attention Mechanism for Deep Reinforcement Learning in Multi-view Environments