分层强化学习原理研究

A Study on the Principle of Hierarchical Reinforcement Learning

首先介绍了强化学习基本原理,分析了马尔科夫决策过程与半马尔科夫决策过程的理论基础及其在强化学习中的应用,其次阐述了分层强化学习中分层与抽象的思想,分析了HAM、Options与MaxQ等方法,并从分层与抽象角度进行了比较,最后指出了分层强化学习的研究发展方向。

Firstly, the principle of RL （reinforcement learning） was introduced and the theories and applications of MDP （Markov decision process） and SMDP for RL were analyzed. Secondly, the concepts of layer division and abstraction were demonstrated and the three HRL methods, including HAM, Options, MaxQ, were analyzed and compared from the aspects of layer division and abstraction. Finally the de- veloping directions of HRL were given.