Towards Learning Scalable Agile Dynamic Motion Planning for Robosoccer Teams with Policy Optimization