摘要
分层强化学习方法可用于解决维数灾难问题,MAXQ方法通过分层地分解值函效,将任务分解为不同层次上的子任务,从而只需在低维空间中解决问题。针对MAXQ方法。首先介绍其基本原理,然后介绍MAXQ方法在出租车问题中的应用,包括任务分解以及类的设计,最后用实验验证了MAXQ方法比Q-学习算法收敛快。
Hierarchical reinforcement learning can be used to solve curse of dimensionality problem. MAXQ method decomposes the task into gubtasks in different levels through decomposing value function hierarchically, so it can be realized in low dimension space. Aiming at the method MAXQ, we firstly introduce the basal principle. Then we introduce the application of the MAXQ method in the taxi problem, including the task decomposition and the class design. Finally,as is testified in practice,MAXQ method converges more faster than Q - learning algorithm.
出处
《茂名学院学报》
2007年第1期56-59,共4页
Journal of Maoming College
关键词
分层强化学习
MAXQ
任务分解
hierarchical reinforcement learning
MAXQ
task decomposition
