Multi-agent learning in the game of guarding a territory

We consider the problem of having a team of guards to learn a joint cooperative strategy to pursue and capture a high speed invader before the invader can reach a territory. In this scenario, the invader is also simultaneously learning its optimal strategy to avoid capture and get as close as possible to the territory. This conflict of interest between the learning agents makes the problem challenging. We adopt the guarding a territory game framework to model the problem, and consider the use of reinforcement learning, particularly the fuzzy actor-critic learning method, to train the players to find their optimal strategies simultaneously. To our knowledge, this is the first work to investigate the development of multi-agent learning for a high speed super invader in the game of guarding a territory. Simulation results from this study demonstrate that all the players are able to learn their optimal behaviors simultaneously