Learning in Large Cooperative Multi-Robot Domains