Toward Human-like Billiard AI Bot Based on Backward Induction and Machine Learning

Abstract

A human-like billiard AI bot approach is proposed in this paper. We analyzed actual game records of human players to obtain feature vectors. The Backward Induction algorithm and machine learning are then proposed to imitate decisions by human players. A run-out sequence is searched backwardly with the assists from heuristics and predictions of neural network models. Through the planning process, a strike unit is found to help guide the physics simulator. With our AI suggestion of strategies, it avoids being over-dependent on the robust and precise physics simulation. Also, we defined an appropriate approach to gauge the human likeness of AI and evaluate our proposed methods. The experimental results show that our method overall is more similar to the way how human players play than that of original AI.