Multi-agent Natural Actor-critic Reinforcement Learning Algorithms