Deep Semantic Acquisition Mechanism Of Robot Service Tasks Based On Environmental Cognition

In a complex and dynamic home environment,the service robot's ability to understand tasks is an important guarantee for whether it can truly understand user needs and provide intelligent services.At present,service tasks understanding mostly adopts the method of extracting key information from task text,but the task text usually contain less information.Robots can only understand shallow semantics.So,they can not correspond the key information in the task texk to the objects in the real environment,which leads to their failure to complete the service tasks smoothly.Therefore,in order to make the robot have a deeper understanding of the service tasks,this paper proposes a deep semantic acquisition mechanism of the robot service tasks based on the environment cognition.Using theoretical knowledge such as image processing,ontology technology and natural language understanding,the environmental cognition is studied,and the semantic knowledge about scenes and objects is obtained.Taking that as the transcendental knowledge,the shallow semantic information of task text is supplemented,so as to obtain the deep semantic information of the service tasks and improve the intelligence level of the robot tasks execution.The main tasks are as follows:Firstly,the semantic knowledge of home environment is extracted accurately and comprehensively based on vision,which can provide deep semantic information for task understanding,such as the scene in which the object is located,the feature of the object and the position relationship with other objects.Knowing the home scene can effectively narrow the scope for the robot to search for objects and improve the service efficiency.The scene category information is obtained by convolution neural network combined with multi-scale coding algorithm.The characteristics of objects and the location relationship with other objects can assist the robots to determine the specific location of the target object in the environment.The FCLN network based on transfer learning is used to extract semantic information such as the appearance characteristics,types and the position relationship of objects.Secondly,the ontology knowledge base of semantic of service environment is constructed to represent the extracted semantic information in a structured and associative way,which solves the problem that the robot cannot understand and use the isolated heterogeneous semantic information directly extracted from the environment,and realizes the efficient storage,management and use of semantic knowledge.Firstly,semantic parser is used to parse the environmental semantic information into a structured form.which can be directly stored in the ontology knowledge base.Then,semantic reasoning is used to mine hidden knowledge and expand the knowledge base.At the same time,the retrieval and modification of the knowledge base are realized,providing an interface for the operation of data for the robot in the task understanding stage.Thirdly,a mechanism based on the semantic semantic ontology knowledge base to obtain the deep semantics of the service task is proposed,which solves the problem of user task semantic missing.On the basis of accurately acquiring the key information of task text,the deep semantics such as scene and feature of objects can be acquired from ontology knowledge base,thus providing necessary semantic knowledge for task planning and execution.Firstly,based on the improved DB-LSTM,the shallow semantic analysis of tasks is carried out to improve the accuracy of shallow semantic analysis.Then using environmental cognitive results as knowledge source,more environmental semantic knowledge needed for task execution is obtained by calling ontology knowledge base query interface and designing its query mode.Finally,the effectiveness of task deep semantic acquisition mechanism in robot task planning and execution is verified.The method proposed in this paper is loaded onto the 3D platform.Combined with the platform task planning and execution module,typical tasks of the robot are planned and executed.It is proved that the method proposed in this paper can effectively improve the intelligent level of robot task execution...