The Study On Chinese Instruction Processing For Home Service Robot

Human-Robot Interaction is an important offshoot of Human-Computer Interaction,and the band of communition between human and robot.As a way of Human-Robot Interaction,Chinese instruction is a form of natural language description of service task.Therefore,it is an important research topic in improving the intelligence and independence of robot that how to transform Chinese instruction into a sequence of actions which robot can understand directly.In this paper,we mainly study on Chinese instruction processing for home service robot,including explicit information extraction,implicit information extraction,people-item relation extraction,and robot task planning.Firstly,aiming at the problem that how to improve the comprehentive of Chinese instruction information extraction at family environment,an explicit information extraction approach based on layered probabilistic hybrid model is proposed.According to the mapping relationship between syntactic structure and semantic information,the definitions of five types of explicit information are predefined.We design the rules of explicit information chunks based on the grammar of Chinese word(or phrase).By referring to the undirected graph thought,a conditional random field model is constructed for chunk labeling.A Bayes classifier is constructed for explicit information extraction using Bayesian statistical method.Through the simulation experiments and explicit information extraction experiments based on Chinese instruction processing system,the viability and effectiveness of proposed model is verified.Secondly,aiming at the problem that,when home service robot understanding Chinese instruction,how to exract the related entity information which is not contained in Chinese instruction,an implicit information extraction approach based on cascaded deep belief networks model is proposed.According to the entity involved in understanding the service task of Chinese instruction,the definitions of two types of implicit information are predefined.By referring to holographic map,the cascaded deep belief networks model is constructed,on the basis of this,the relevant implicit information is extracted using energy function.Through the simulation experiments and implicit information extraction experiments based on Chinese instruction processing system,the viability and effectiveness of proposed model is verified.Thirdly,aiming at the problem that the existing semantic relation extraction approaches can not directly transplanted to extract many types of semantic relation between people and item at family environment,a people-item relation extraction approach based on multiple kernel support vector machine model is proposed.According to the connection between people and item in home environment,the definitions of four types of people-item relation are predefined.By referring to the lexical feature,phrase feature,order feature and property feature of entity in Chinese instruction,the multi-features combination of entity pair is constructed.Moreover,the lexical kernel,syntactic kernel and property kernel as the basis kernel functions,multiple kernel function is constructed using the weighted sum method.On the basis of that,the simple multiple kernel learning algorithm is used to construct the multiple kernel support vector machine model.Through the simulation experiments and people-item relation extraction experiments based on Chinese instruction processing system,the viability and effectiveness of proposed model is verified.Finally,aiming at the problem that the existing task planning approaches can not parse Chinese instruction into a sequence of actions which is understood by home service robot,a robot task planning approach based on parse template is proposed.According to the number and variety of entity in implicit information,the types of Chinese instruction and the definitions of relevant service task are given.By referring to the people-item relation and task/step joined table,the different parse templates are designed for different service tasks.We design the action/status transfer process of robot using the task decomposition rule,on the basis of this,a sequence of actions that correspond to Chinese instruction can be obtained.Through the simulation experiments and robot task planning experiments based on Chinese instruction processing system,the viability and effectiveness of proposed parse template is verified.