Learning to Simulate Self-driven Particles System with Coordinated Policy Optimization