Cooperative Cognitive Agents and Reinforcement Learning in Pursuit Game

This paper illustrates how a self-organizing cognitive architecture, known as TD-FALCON, can learn to function and cooperate in a dynamic environment. TD-FALCON learns the value functions of the stateaction space estimated through a temporal difference (TD) method. The learned value functions are then used to determine the optimal actions based on an action selection policy. To tackle a multi-agent predator/prey pursuit task, we develop a cooperative strategy using a high-level compressed state representation and a hybrid reward function. Experiments show that TD-FALCON agent teams operating with the proposed cooperative strategy produce superior performance with a high level of efficiency and scalability.