Apply Deep Reinforcement Learning to NS-SHAFT Game Control
Bo-Yu Lin, Ching-Lung Chang, Chuan‐Yu Chang · 2020
Reinforcement Learning (RL) with both exploration and exploit abilities is applied to games. Related literature shows that it can surpass human performance, and some can even defeat human. This paper is mainly based on the combination of reinforcement learning and deep learning, called Deep Q-Network, to learn the action response of game NS-SHAFT autonomously. Based on a personal computer, we built an experimental learning environment that automatically captures the NS-SHAFT's frame which is provided to DQN to decide the action of moving left, moving right, or stay in same location, survey different parameters: such as the learning cycle, different reward settings, and exploration / exploit ratios etc., which affect the learning effectiveness. The experimental results show that moderate parameter settings have a certain degree of impact on the DQN learning effect.