基于Q-Learning的智能体训练

How to Train the Agent Based on the Way of Q-Learning

针对机器人足球比赛的多智能体环境下智能体的训练问题,提出了一种将模糊控制与Q-Learning相结合的学习方法,并在学习过程中自动调节回报函数以获得最优策略,此方法的有效性在中型组的仿真平台上得到了验证,并取得了较好效果,还可将它改进应用于其他多智体环境。

In order to solve the problem of agent training in the multi-agent circumstances of robot soccer, a new method of agent learning is put forward, which combines the fuzzy control with the Q-Learning. During the learning process, the reward function is controlled automatically to earn the optimal policy. It is proved that this method is effective on the simulation platform of middle-size league, and the simulation result is good. In addition, it could be adapted to apply in other multi-agent circumstances.