A curriculum-based reinforcement learninig approach to pedestrian simulation