基于MDP的移动机器人室内导航对话管理实现

摘要

A spoken dialog system for the navigation of indoor mobile robots is proposed as a solution for improving the understanding of natural spoken instructions for the navigation of remote mobile robots.The design of the dialog management (DM)system is based on the model of Markov Decision Process(MDP).The states of MDP are the set of combination of the navigation elements with different property values from the spoken navigation instructions,and the actions are the set of re-actions from the dialogue system.The training result of the MDP model is that the robots should select a proper action corre-sponding to the state recognized from the elements of navigation instruction given by the human operator,which should make a natural,fluent dialog with the eventual purpose of consensus understanding of the instruction between human and the robot.The natural language generation of the dialog system is based on Artificial Intelligence Markup Language(AIML).Ex-periments show that the dialogue system works naturally and fluently with the accurate rate of 85.16 %.%针对自然语言指令对远程移动机器人导航控制中自然语言理解达不到要求这一问题,提出通过人机对话提高机器人对自然语言指令理解的程度.设计了一个面向救灾机器人室内导航的对话管理系统,建立了对话管理系统的马尔科夫决策过程(MDP)模型.该模型将自然语言指令中导航要素及其属性的组合定义为状态集合,把机器人作出的对话回应定义为动作集合.模型训练的结果是从导航指令中识别出具体的状态,机器人能作出合理的回应,以使对话能够自然流畅地进行,达到最终正确理解导航指令.机器人的自然语言生成基于人工智能标记语言(AIM L)设计.实验测试表明,对话过程流畅自然,用户体验较好,准确率达到85.16%.