Research On Service Task-Oriented Robot Semantic Knowledge Assisted Object Cognition Mechanism

In recent years,research on service robots have achieved remarkable devel-opment.However,service robots still focus on simple functions such as cleaning,escort,education and entertaining.Advanced service task(like food processing,tableware arrangement and so on)remains absence.Insufficient cognitive ability to objects has become a key bottleneck problem that restricts the development of service robots.Such as service robots' perception of objects mostly focus on de-tection.Aiming at the classification,target location and correlation.This cognitive skill cannot meet the requirements for further understanding of objectives,opera-tional objectives,and provision of high-level services.Humans can use semantic knowledge to assist the recognition of target objects.Including target semantic de-scription,rule reasoning,inter-target association,and guided target search.Aiming at the problem of insufficient cognitive ability of service robots to target objects.In order to mimic using semantic knowledge to assist cognitive mechanism,this pa-per conducted a study on semantic knowledge based service robot cognition in the domestic environment.Such as semantic knowledge modeling,reasoning,supple-mentation,guided detection and related goals.The purpose is to enable robots to grasp the target object by designing efficient semantic knowledge,break through the cognitive bottleneck of the service robot,and improve the intelligence degree of the service robot.So as to realize the relatively complex service task operation function of the robot.Therefore,this paper focus on the semantic cognition mechanism of target objects of service robots.This paper has important theoretical significance and practical application value.With the development of research on distributed sensing,deep neural network and semantic analysis.Research on semantic processing and visual perception for robots has also made significant progress.Some aspects have reached or exceeded the human level.However,there is a lack of research work that combines seman-tic knowledge with visual perception in order to assist robots in cognition.It is necessary to comprehensively use relevant approaches to construct an artificial neu-ral network for detecting objects,as well as a knowledge base for storing semantic knowledge.Based on this,semantic knowledge description,intelligent spatial selec-tive attention,semantic knowledge association discovery,semantic auxiliary context reasoning and recognition Research on enhancement and other aspects to achieve the auxiliary knowledge of semantic knowledge on the cognitive process of service robots in the domestic environment.This paper studies six main aspects involved in semantic knowledge-assisted robot cognitive process.Specific research contents and research results include the following aspects:(1)For the semantic knowledge support need of the service robot target object cognitive process,a semantic knowledge modeling method for environment-robot-service triple structure is proposed.Firstly,the ontology-based knowledge represen-tation method is designed to simulate the hierarchical structure of human semantic knowledge.The semantic-knowledge description method of the environment-robot-service task triple is designed as the core architecture,and the corresponding seman-tic knowledge base is built.A hierarchical representation of the semantic knowledge of cognitive processes in service robots.Secondly,based on the description logic,the semantic rules and reasoning mechanism are designed.The inference rules based on description logic and the corresponding query inference engine are constructed to realize the reasoning of hierarchical entities and associations such as class relations and instance attributes.The example verification shows that the method is oriented to the top-level design of the service task execution process,and solves the problem of knowledge representation,memory and query reasoning of the cognitive process of supporting service robots,and realizes the deep correlation and fusion of deep knowledge and service execution of goods.(2)For the needs of locating FOA associated with service tasks in the domes-tic environment.A selective attention mechanism based on intelligent space is de-signed.Firstly,dynamic features and static features are extracted at multiple scales based on the visual information.The event items are used to represent the discrete sensing information under the intelligent space platform.Multiple distinctive fea-tures are integrated to describe the spatial saliency.Then,a plurality of feature maps are merged to form a saliency map based on the information entropy calculation.The switching strategy for FOA is performed.Experiments show that this method can effectively locate the FOA according to service tasks in intelligent space.(3)Aiming at the problem that the detection of objects in the domestic envi-ronment lacks a priori guidance.As well as the detection efficiency is low and the detection results are relatively simple.A method for detecting and describing the objects guided by selective attention is proposed.Firstly,an object detection method based on regional convolutional neural network under selective attention guidance is constructed.The region extraction process of regional convolutional neural network is guided by selective attention mechanism.The objects related to service task are selected.Then,the scene graph generation method based on semantic analysis is proposed.The knowledge stored in the semantic knowledge base is used to describe the details of the scene.Experiments show that this method can improve the effi-ciency of object detection,as well as generate a scene graph description with rich semantics.(4)Based on the requirements of semantic knowledge association mining and context reasoning.A semantic relation mining method based on semantic analy-sis and a context inference mechanism based on petri net is constructed.Firstly,the latent semantic analysis model is designed to extract topic models.The topic models contain instruction of service task execution from the Internet.The apriori algorithm is used to obtain the association relationship between the service tasks and the required objects in the topic model.The semantic association is generated by using the Internet semantic description documents.A context inference mechanis-m based on improved fuzzy petri net is proposed.A discrete dynamic system for service-oriented scenes is constructed.A fuzzy petri net and corresponding context inference mechanism for semantic knowledge is proposed.Experiments show that the mechanism can mine the semantic association of the massive service task de-scription document in the Internet and realize the context reasoning supported by the semantic association.(5)In order to meet the need for cognitive enhancement through learning expe-rience.A cognitive enhancement mechanism combining instance transfer learning and semantic knowledge complementation is proposed.Firstly,an improved TrAd-aBoost algorithm for instance migration is proposed.The TrAdaBoost method is improved by combining time factor.The cognitive process instance is filtered and weighted.The migration weighted instance is used to train the convolutional neural network to realize instance-based migration.The classification ability of the object is improved.Then,the semantic knowledge base complement method is designed.The semantic knowledge is mapped to the low-dimensional space.The analogy is utilized to infer the hidden relationship between the entities,and the missing knowl-edge between the entities in the knowledge base is completed.Experiments show that this method can effectively select and migrate instances to enhance the neural network classification ability,and implement the semantic knowledge base comple-mentation for the missing semantics of analogical reasoning.