A discussion on planning method in platform game AI -A study of situation recognition for utilizing learning data-

Abstract

Reinforcement learning and search algorithm are typical methods to realize autonomous agent. However, the agent by these methods may not obtain action rules autonomously because of state space and constraint condition. We proposed the method which combine A* algorithm and Q-learning in our previous study. This method enable agents to obtain action rules autonomously at the environment which was not possible with previous method. However, the learning data is optimized to the learning environment. Therefore, it is difficult to utilize the learning data in different environment. In this article, we target platform game benchmark and discuss the method which utilizes the learning data in different environment.