Research On Robot Grasping System Based On SSD Object Detection Technology

With the proposal of the "Made in China 2025" power strategy,robots have received more and more attention in the fields of industry,agriculture and military industry.In the field of industrial manufacturing,robotic automation can reduce the labor intensity of workers,increase production efficiency and improve product quality.At present,industrial robots mainly perform coding through teaching methods.The robots perform prescribed actions according to established procedures.The advantage is that the programming method is simple and efficient.The disadvantage is that it cannot cope with complicated and changeable working environments.In the industrial production process,a single robot grasps a single type of component,and the component needs to be placed in a fixed position,otherwise the robot cannot complete the grasping task.However,in the human-robot interaction work environment,a variety of components are mixed and stored,and the placement position is not fixed or may be changed at any time.Using the traditional teaching method to program the robot obviously cannot complete the task.In order to solve the problem of mixed placement and unfixed position of various objects,this paper will apply SSD object detection technology to the robot grasping system.Firstly,collect a large number of object images for training our SSD object detection model.Then,based on the SSD model,we can get the bounding box of the target object in the image.Determine the pixel coordinate and depth value of the object from the position of the bounding box and then calibrate with the Kinect camera.Then,the camera's hand-eye calibration technique is used to map pixel coordinates and depth values into the robot's base coordinate system.Finally,the robot moves and grasps the object according to the position of the object in the base coordinate system.This paper mainly completed the following works:1.Collect object images at different positions and postures on the workbench and then augment the dataset with data enhancement techniques.The data set is used as training samples to train the SSD object detection model,and the model that meets the requirements is finally obtained.2.The D-H table of the robot is obtained according to the robot joint size information,and the homogeneous transformation matrix of adjacent joints is calculated according to the D-H table,and inverse kinematics is solved to obtain the mapping relationship between the end pose of the robot and the joint space.3.The 9×6 calibration plate was fixed on the end of the robot,and 20 calibration plate images and corresponding joint angle values in different positions and postures were collected.Using the Zhang Zhengyou calibration method,the Kinect camera was calibrated,and the camera internal parameter K and the translation vector and rotation vector of each calibration plate in the camera coordinate system were obtained.Calculate the homogeneous transformation matrix from the camera coordinate system to the robot base coordinate system based on the robot kinematics solution.4.Use the SSD model to detect the image captured by Kinect to obtain the pixel coordinates and depth values of the target object in the image.According to the hand-eye calibration results,the pixel coordinates and depth values are mapped to the robot base coordinate system,and the position coordinates of the object in the base coordinate system are obtained.Finally,the position coordinates of the object are sent to the robot,and the robot performs the object's grasping task.