项目名称: 基于事件的强化学习及其在群机器人优化控制中的应用
项目编号: No.61273327
项目类型: 面上项目
立项/批准年度: 2013
项目学科: 自动化技术、计算机技术
项目作者: 陈春林
作者单位: 南京大学
项目金额: 79万元
中文摘要: 强化学习是实现智能系统的一项关键技术,利用问题的结构信息设计分层学习算法是克服复杂问题维数灾难、提高学习速度的有效方法。本项目拟以实际群机器人系统中的大规模空间优化控制问题为背景,将基于事件的优化方法引入到强化学习系统的设计与分析中,系统性的提出基于事件的强化学习(Event-based Reinforcement Learning, ERL)方法,以有效解决模型未知的非标准马尔科夫决策问题,并深入研究其理论、算法及典型应用,包括:①基于对事件的形式化表示,研究ERL基本模型和算法理论;②针对实际工程需求,提出基于概率模糊系统的事件表示和推理方法,研究实用的ERL快速迭代算法;③结合ERL理论方法创新,研究ERL在群机器人大规模网络优化及协调控制中的应用。本研究对探索具有结构特征和推理能力的强化学习理论和算法具有重要理论价值和现实意义,也将促进ERL在实际大规模空间优化控制问题中的应用。
中文关键词: 强化学习;基于事件的学习;群机器人;量子鲁棒控制;优化控制
英文摘要: Reinforcement learning is one of the key techniques to implement intelligent systems. It is an effective way to overcome the curse of dimensionality and speed up learning by designing hierarchical learning algorithms using the structural information of the problems. This project will take the large scale optimal control problems in the real swarm robot systems as the background of applications. Aiming at solving nonstandard Markov decision problems with unknown models, the event-based optimization method is adopted for the design and analysis of reinforcement learing system and an event-based reinforcement learing (ERL) method will be systematically proposed. Then the theories, algorithms and typical applications of ERL will be comprehensively studied. The main research contents include the following three aspects: (1) based on the formalization of events, research on the fundamental models and algorithmic theories of ERL; (2) regarding the needs of practical engineering, propose a probabilistic fuzzy system based representation and reasoning methods for events, then focus on the study of fast iterative algorithms of ERL; (3) research on the applications of ERL in the large scale network optimization and coordination control of swarm robots. This project will be very important for the exploration of RL theories
英文关键词: Reinforcement learning;Event-based learning;Swarm robots;Quantum robust control;Optimal control