摘要
已有的动作模型学习方法针对确定的或不确定的瞬时动作,而未考虑动作模型中的时态关系。提出了在部分观测环境下自动学习时态动作模型的方法。设计了学习动作持续时间表达式一般形式的两阶段线性回归方法。通过分析命题时间戳设计了动作前提、效果与动作之间时态关系算子的构建算法。在"国际智能规划竞赛"的规划问题集上进行了实验,结果表明了该方法的有效性。
There exist several successful action model learning methods for instaneous actions with certain or uncertain effects. However, few of them consider the temporal relationships underlying domain models for more realistic environments. The paper proposes a method for the automatically learning of temporal action model in the settings of partially observable environment. The general form of a duration's expression is learnt by a two-stage linear regression method.The temporal relationships among preconditions, effects and operators are constructed by analysis on propositions' timestamps in plan traces. Experiments on the temporal planning benchmarks from the International Planning Competition(IPC)are carried out and the results show that the method is effective.
出处
《计算机工程与应用》
CSCD
北大核心
2018年第2期29-39,共11页
Computer Engineering and Applications
基金
国家自然科学基金(No.61103136)
武汉工程大学研究生教育创新基金(No.CX2016061)
关键词
智能规划
动作模型学习
时态关系
线性回归
领域建模
知识获取
automated planning
action model learning
temporal relation
linear regression
domain modeling
knowl edge acquisition