Research On Fast Search Strategy For Objects For Robots Service Tasks

The service tasks performed by the robot in the indoor environment(such as taking books,water for the owner),require the robot to perform path planning according to the target semantics and search for the target in a timely and accurately.However,when working in a wide range of workspaces,the objects related to the tasks may exceed the operable range of the robot.Therefore,the robot must search for objects in the environment before performing the task.When the object to be searched is small and the position is not fixed,it will increase the difficulty and time of the searching and cannot meet the service needs of people.Therefore,the research of quick search of objects in the indoor environment has become the key and prerequisite for the service robot to complete the service task.The project design simulates people's hierarchical perception search of the environment,and constructs a three-level attribution chain of belonging objects-identification objects-function room,constitutes the environment description form consistent with the human space memory habit,and realizes the intelligence of the three-level hierarchical semantic path planning.At the same time,the Markov probability model is constructed for semantic missing or invalid in the attribution chain,and realize the probability description of the position of indoor objects by the service robot.(1)Construct the identification-belonging relationship.The target objects(belonging objects)that the robot needs to search for when completing the intelligent service task is generally small in size,not fixed in position and usually placed on the object(identification objects)with the support plane,thus imitating the space search mode of the person and combining the position structure information between the objects,it is proposed to optimize the search efficiency of the belonging objects by using the identification objects.Using the ground convex hull information as the criterion,the indoor objects are divided into the identification objects and the belonging objects.The CVFH(Cluster Viewpoint Feature Histogram)feature is used for recognition to form the identification library and the attribution library.At the same time,associating the belonging objects owned by the identification objects to construct the identification-belonging relationship,and the anthropomorphic search process is implemented according to the identification-belonging relationship.(2)Dividing room function area.The specific objects searched by the service robot must be related to the function of the room in which it is located.It is necessary to establish a room concept for the environmental space to determine the attribution relationship between the target object and the room,and to narrow the search space according to the target semantic information;The continuous environmental space is discretized into multiple functional rooms,connected by the topology paths,to realize the semantic path planning during the search.Therefore,starting from the requirements of path planning and search efficiency,constructing a topology search path with functional rooms as nodes.Combining visual features and geometric features to determine the connectivity relationship between the observed nodes to form an undirected weighted graph.Then the group division is performed based on the convolutional neural network.According to whether there is a communication node between the subgroups,the inter-group communication relationship is determined.And constructing an indoor topology map with each functional room as a node.Associating identification-belonging relationship in each room with room function to build a three-level association relationship to achieve three-level hierarchical search,and optimize search efficiency.(3)Constructing Markov chain probability model.For semantic missing or invalid items in the attribution chain,it is difficult to achieve fast search because of the lack of a fixed three-level association relationship.Such objects have a certain degree of semantic correlation and spatial correlation with the function of the room in which they are located.Therefore,the concept of divide and conquer is used to develop different search strategies.Based on the idea of greedy algorithm,the semantic mapping model is used to initialize the association probability between the belonging objects with the function room,identification objects.Combining with the physical influence factors,the room-to-room and in-room,search strategies are initialized.Then the three-level attribution chain of belonging objects is learned through multiple search cycles,and building a Markov probability model to inversely optimize search strategy to achieve a probabilistic description of the location of the objects.The method proposed in this paper closely combines the environmental cognitive form of the service robot with the human mindset,improves the interaction ability between the robot and the environment,realizes the visual perception and hierarchical awareness of the robot to the space environment,and lays a foundation for the robot to better provide intelligent services.In this paper,the method is verified by experiments,which proves the effectiveness of the method and has important theoretical and practical value for realizing robot intelligent service.