项目名称: 基于Markov逻辑网络的HTN模型获取算法
项目编号: No.61309011
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 自动化技术、计算机技术
项目作者: 卓汉逵
作者单位: 中山大学
项目金额: 27万元
中文摘要: 在HTN(分层任务网络)规划中,人工建立HTN模型往往是很耗时和困难的,特别是当各种逻辑关系很复杂时。这已经成为应用HTN规划技术解决实际问题的瓶颈。本项目旨在研究如何从历史数据中自动学习获得HTN模型,以降低人工建立HTN模型的耗费,从而促进推广HTN规划技术的应用。本项目首先建立一组候选逻辑公式,用以描述各种可能的HTN模型;然后,将历史数据转化为逻辑命题形式,用以学习候选逻辑公式的权重;最后,改进Markov逻辑网络学习算法,并利用改进的算法学习得到HTN模型。在实验中,本项目在不同的HTN规划领域中验证算法的有效性和高效性。
中文关键词: 人工智能;智能规划;分层任务网络;Markov 逻辑网络;
英文摘要: In HTN (Hierarchical Task Network) planning, creating HTN models by hand is both time-consuming and difficult, especially when relations among objects are very complicated. This has become the bottleneck of applying HTN planning techniques to solve real application problems. In this project, we aim to study how to automatically learn HTN models from history data, such that human effort of building HTN models is reduced and HTN planning techniques are applied more widely. We first build a set of candidate logical formulae to describe all possible HTN models. After that, we transform history data into a set of propositions for learning the weights of candidate logical formulae. We then change the learning algorithm of Markov Logic Networks (MLNs) to finally learn the HTN models. In the experiment, we evaluate the effectiveness and efficiency of our algorithm in different HTN planning domains.
英文关键词: Artificial Intelligence;AI Planning;Hierarchical Task Network;Markov Logic Networks;