Shaping AI Behavior: A Q-Learning Driven Approach to Automatic Behavior Tree Creation

In this paper we propose a novel solution to generate Behavior Trees automatically and autonomously from reinforcement learned autonomous agents. We developed a Capture The Flag-style game in which two teams compete to win. The Behavior Trees are generated from the knowledge of Q-Learned autonomous agents competing in this game by extracting their knowledge and parsing it into a Behavior Tree format. The proposed algorithm can generate these trees with comparable performance to the autonomous agents as well as compete with handmade state machine like solutions.