| With the rapid development of artificial intelligence and the popularization of robots,robots play an important role in industrial production and family life.Manmachine hand-over tasks are widely used in various scenarios,such as robots assisting workers in factories,serving family members and disabled people in families.In these scenarios,not only efficient and accurate delivery should be achieved,but also the safety of the interactive personnel should be ensured.This problem involves the perception and processing of target detection and external information acquisition,which is an important challenge for researchers.However,the current man-machine hand-over tasks mainly have problems such as complex system construction,high equipment requirements,and insufficient attention to the safety of interactive personnel.In order to enable robots to deliver users efficiently,smoothly and safely in unstructured scenes,this paper designs and builds a human-robot hand-over system in unstructured scenes.Firstly,the Kinect camera is used to obtain the color and depth information in the scene,and then the image is sent to the trained YOLOv5 rotating target network to obtain the pose of the object in the human hand in the camera coordinate.Then,the camera calibration is converted to the world coordinate system.After that,the position tracking of the object in the human hand is realized by the proposed servo control strategy.Finally,the flexible hand-over between the human hand and the robot is realized by impedance control in the interaction stage.The main work of this paper is as follows:(1)In this paper,the YOLOv5 dataset of objects in human hands is produced in the experimental scene,and then the dataset is trained in the built YOLOv5 network framework,and the performance of the network model is evaluated.The problem of pose recognition of objects in human hands in unstructured scenes is solved.At the same time,the position accuracy of objects in human hands is identified and evaluated in the experimental scene,which verifies that the position detection accuracy meets the requirements of human-robot hand-over.(2)In order to obtain the representation of the object in the world coordinate system,the calibration between the camera and the robot and the pose conversion of the recognition results are carried out in this paper.At the same time,in order to make the robot track the identified position,this paper proposes a dynamic step size adjustment strategy based on position error,which solves the problems of the stability of the robot servo and the security of the operator in the human-robot hand-over delivery process.The position error of the experimental scene and the motion trajectory of the claw are analyzed to verify the effectiveness of the servo control.(3)This paper builds a human-robot hand-over delivery platform,and the server builds a deep learning framework under Ubuntu,and sends the identification results to the client through TCP communication.The client uses the delivery control algorithm proposed in this paper to realize the whole process.By introducing impedance control model,the problem of flexible delivery of objects in the process of man-machine delivery is solved.At the same time,experiments and data analysis are carried out on the whole process.The experimental results show that the human-robot hand-over delivery system proposed in this paper can deliver efficiently,safely and smoothly in unstructured scenes. |
