نوع مقاله : مقاله پژوهشی
نویسندگان
1 دانشگاه صنعتی امیرکبیر
2 دانشگاه صنعتی امیرکبیر.
چکیده
کلیدواژهها
عنوان مقاله [English]
نویسندگان [English]
In this paper, multiagent reinforcement learning has been used in wargame simulation. Military organization causes priorities in actions taken among agents in the feild. Decision making among n-person generalsum multiagent systems with consecutive action selection can be modeled through extensive form games. The existing process of learning can be considered as a set of successive extensive form games with Markov property. We use subgame perfect equilibrium points as optimal joint action using the well-known backward induction algorithm. An estimation of possible utilities gained over a game with respect to other agents’ preferences, called as associative Q-values, is used to predict optimal action. Simulation results present the effectiveness of the proposed algorithms in wargame simulation.
کلیدواژهها [English]
)