State Space Segmentation for Acquisition of Agent Behavior

We propose a new method to categorize continuous numeric percepts for Q-learning, where percept vectors are classified into categories and Q-learning uses categories as states to acquire rules for agent behavior. In the proposed method, categories are represented by hyper-spheres. A percept vector is classified into a category that covers the vector and is the nearest from it. For efficient reinforcement learning, category merging is provided with the method, where the number of parameters to control category merging in the method is fewer than that in modified fuzzy ART. The proposed method is combined with Q- learning and it is compared with Q-learning with original and modified fuzzy ART. Experimental results show that our method learns good rules for agent behavior more efficiently than Q-learning with modified fuzzy ART.