机器人自然语言导航的层叠式条件随机场模型

提出了一种基于层叠条件随机场进行救灾机器人自然语言导航命令理解的方法。该方法由三层条件随机场(CRFs)构成:第一层用于导航词性标注,选取词、词性以及上下文作为特征模板生成导航词性标签;第二层用于导航过程提取,选择词、导航词性标签以及上下文构建特征模板生成导航过程标签;第三层用于起点终点识别,选取词、导航词性标签、导航过程标签以及上下文构建特征模板判断出地名词为起点还是终点。根据导航词性与导航要素的对应关系便可从命令中提取出导航信息。该方法能够处理完全不受限的自然语言导航命令,总体正确率达到78.6%,无需依赖特定的指令与地图,对完成救灾机器人导航的人机交互任务具有重要意义。

A cascaded conditional random fields model of natural language processing for the navigation of rescue robots

We propose a new method for rescue robots to understand navigation commands in Chinese natural language based on cascaded conditional random fields (CRFs). It consists of three layers of CRFs. The first layer is to tag the navigation part of speech (NPOS) using features from words, parts of speech and the context. The second layer is to extract basic navigation procedures (NPs) using features from words, NPOS labels and the context. The third layer is to recognize start places and end places of each NP using features from words, NPOS labels, NP labels and the context. Eventually, according to the relationship between the NPOSs and navigation elements, navigation information can be obtained from the navigation commands. The method can process navigation commands of uncontrolled natural language and the accuracy is 78.6%. It does not depend on custom-made instructions or maps, which is significant for rescue robot navigation through human-robot interaction.