Centralized Model and Exploration Policy for Multi-Agent RL