Not all agents are equal: scaling up distributed POMDPs for agent networks