Efficient Inter-Team Task Allocation in RoboCup Rescue

The coordination of cooperative agents involved in rescue missions is an important open research problem. We consider the RoboCup Rescue Simulation (RCS) challenge, where teams of agents perform urban rescue operations. Previous approaches typically cast such problem as separate single-team allocation problems. However, different teams have complementary capabilities, and therefore some kind of inter-team coordination is desirable for high-quality solutions. Our contribution considers inter-team coordination using Max-Sum. We present a methodology that allows teams in RCS to efficiently assess joint allocations. Furthermore, we show how to reduce the algorithm's computational complexity from exponential to polynomial time by using Tractable High Order Potentials. To the best of our knowledge this is the first time where it has been shown that MS can be run in polynomial time in the RCS challenge without relaxing the problem. Experiments with fire brigades and police agents show that teams employing inter-team coordination are significantly more effective than uncoordinated teams. Moreover, the evaluation shows that our BMS and THOPs method achieves up to $2.5$ times better results than other state-of-the-art methods.