Simulating Emergent Properties of Human Driving Behavior Using Multi-Agent Reward Augmented Imitation Learning