论文部分内容阅读
为了解决Q学习应用于连续状态空间的智能系统所面临的“维数灾难”问题,提出一种基于ART2的Q学习算法.通过引入ART2神经网络,让Q学习Agent针对任务学习一个适当的增量式的状态空间模式聚类,使Agent无需任何先验知识,即可在未知环境中进行行为决策和状态空间模式聚类两层在线学习,通过与环境交互来不断改进控制策略,从而提高学习精度.仿真实验表明,使用ARTQL算法的移动机器人能通过与环境交互学习来不断提高导航性能.
In order to solve the problem of “dimensionality disaster ” faced by Q learning intelligent system applied to continuous state space, a Q learning algorithm based on ART2 is proposed.By introducing ART2 neural network, Q learning agent can learn a proper Incremental clustering of state-space patterns enables the Agent to conduct two-tier online learning of behavioral decision-making and state-space pattern clustering in an unknown environment without any prior knowledge, and improves the control strategy continuously by interacting with the environment Learning accuracy.The simulation results show that the mobile robot using ARTQL algorithm can improve the navigation performance by interacting with the environment.