Recent Advances in Hierarchical Reinforcement Learning

Reinforcement learning is bedeviled by the curse of dimensionality: the number of parameters to be learned grows exponentially with the size of any compact encoding of a state. Recent attempts to combat the curse of dimensionality have turned to principled ways of exploiting temporal abstraction, where decisions are not required at each step, but rather invoke the execution of temporally-extended activities which follow their own policies until termination. This leads naturally to hierarchical control architectures and associated learning algorithms. We review several approaches to temporal abstraction and hierarchical organization that machine learning researchers have recently developed. Common to these approaches is a reliance on the theory of semi-Markov decision processes, which we emphasize in our review. We then discuss extensions of these ideas to concurrent activities, multiagent coordination, and hierarchical memory for addressing partial observability. Concluding remarks address open challenges facing the further development of reinforcement learning in a hierarchical setting.     

分层强化学习中的动态分层方法研究

分层强化学习中现有的自动分层方法均是在对状态空间进行一定程度探测之后一次性生成层次结构,不充分探测不能保证求解质量,过度探测则影响学习速度,为了克服学习算法性能高度依赖于状态空间探测程度这个问题,本文提出一种动态分层方法,该方法将免疫聚类及二次应答机制融入Sutton提出的Option分层强化学习框架,能对Option状态空间进行动态调整,并沿着学习轨迹动态生成Option内部策略,以二维有障碍栅格空间内两点间最短路径规划为学习任务进行了仿真实验,结果表明,动态分层方法对状态空间探测程度的依赖性很小,动态分层方法更适用于解决大规模强化学习问题.     

分层强化学习综述

近年来强化学习愈发体现其强大的学习能力,2017年AlphaGo在围棋上击败世界冠军,同时在复杂竞技游戏星际争霸2和DOTA2中人类的顶尖战队也败于AI之手,但其自身又存在着自身的弱点,在不断的发展中瓶颈逐渐出现.分层强化学习因为能够解决其维数灾难问题,使得其在环境更为复杂,动作空间更大的环境中表现出更加优异的处理能力...     

分层强化学习中的Option自动生成算法

分层强化学习中目前有Option、HAM和MAXQ三种主要方法,其自动分层问题均未得到有效解决,该文针对第一种方法,提出了Option自动生成算法,该算法以Agent在学习初始阶段探测到的状态空间为输入,采用人工免疫网络技术对其进行聚类,在聚类后的各状态子集上通过经验回放学习产生内部策略集,从而生成Option,仿真实验验证了该算法的有效性。     

分层强化学习研究综述

强化学习通过试错与环境交互获得策略的改进,其自学习和在线学习的特点使其成为机器学习研究的一个重要分支.但是,强化学习一直被"维数灾"问题所困扰.近年来,分层强化学习方法引入抽象(Abstraction)机制,在克服"维数灾"方面取得了显著进展.作为理论基础,本文首先介绍了强化学习的基本原理及基于半马氏过程的Q-学习算法.然后介绍了3种典型的单Agent分层强化学习方法(Option、HAM和MAXQ)的基本思想,Q-学习更新公式,概括了各方法的本质特征,并对这3种方法进行了对比分析评价.最后指出了将单Agent分层强化学习方法拓展到多Agent分层强化学习时需要解决的问题.     

基于MAXQ方法的分层强化学习

强化学习是机器学习领域的一个重要分支,但在强化学习系统中,学习的数量会随着状态变量的个数成指数级增长,从而形成"维数灾".为此提出了一种基于MAXQ的分层强化学习方法,通过引入抽象机制将强化学习任务分解到不同层次上来分别实现,使得每层上的学习任务仅需在较小的空间中进行,从而大大减少了学习的数量和规模.并给出具体算法--MAXQ-RLA.     

分层增强学习在足球机器人比赛中的应用

足球机器人的研究是一项挑战性的研究领域,为了设计出智能型的球员必须涉及到计算机、人工智能、视觉及机械学等方面的研究。球员的学习能力是体现其智能的主要标志。如何在不断改变的外界环境中选取合适的动作技巧是在机器人足球比赛中的一个关键问题。该文介绍了马尔可夫决策过程,在半马尔可夫决策模型下,利用分层增强学习算法对不同层次的动作学习和选取同时进行学习。在仿真平台上进行实验,结果表明该学习方法是非常有效的。     

基于路径匹配的在线分层强化学习方法

如何在线找到正确的子目标是基于option的分层强化学习的关键问题.通过分析学习主体在子目标处的动作,发现了子目标的有效动作受限的特性,进而将寻找子目标的问题转化为寻找路径中最匹配的动作受限状态.针对网格学习环境,提出了单向值方法表示子目标的有效动作受限特性和基于此方法的option自动发现算法.实验表明,基于单向值方法产生的option能够显著加快Q学习算法,也进一步分析了option产生的时机和大小对Q学习算法性能的影响.     

Hierarchical Reinforcement Learning With Automatic Sub-Goal Identification

In reinforcement learning an agent may explore ineffectively when dealing with sparse reward tasks where finding a reward point is difficult. To solve the problem, we propose an algorithm called hierarchical deep reinforcement learning with automatic sub-goal identification via computer vision (HADS) which takes advantage of hierarchical reinforcement learning to alleviate the sparse reward problem and improve efficiency of exploration by utilizing a sub-goal mechanism. HADS uses a computer vision method to identify sub-goals automatically for hierarchical deep reinforcement learning. Due to the fact that not all sub-goal points are reachable, a mechanism is proposed to remove unreachable sub-goal points so as to further improve the performance of the algorithm. HADS involves contour recognition to identify sub-goals from the state image where some salient states in the state image may be recognized as sub-goals, while those that are not will be removed based on prior knowledge. Our experiments verified the effect of the algorithm.      

When Does Communication Learning Need Hierarchical Multi-Agent Deep Reinforcement Learning

Abstract

Multi-agent systems need to communicate to coordinate a shared task. We show that a recurrent neural network (RNN) can learn a communication protocol for coordination, even if the actions to coordinate are performed steps after the communication phase. We show that a separation of tasks with different temporal scale is necessary for successful learning. We contribute a hierarchical deep reinforcement learning model for multi-agent systems that separates the communication and coordination task from the action picking through a hierarchical policy. We further on show, that a separation of concerns in communication is beneficial but not necessary. As a testbed, we propose the Dungeon Lever Game and we extend the Differentiable Inter-Agent Learning (DIAL) framework. We present and compare results from different model variations on the Dungeon Lever Game.