| Traditional robots can be well programmed to perform simple routes and fixed movements in industrial production,which greatly liberates social productivity.With the improvement of technology,people have more expectation for the diversity of robot functions.Especially with the development of image processing technology and machine learning methods,robots have become more intelligent and functionally richer.In order to enhance the robot’s autonomy and improve its resilience and processing ability in complex scenes,this paper combines machine vision,robot modeling,motion analysis and motion planning technologies to build a panda robot item grasping system in complex scenes.Firstly,this paper proposes the overall design scheme of vision-based robot grasping system,and selects appropriate hardware and software to build the overall framework.Secondly,we model the panda robot by D-H parameter method and derive the urdf model to analyze and solve the positive and negative kinematic problems based on the characteristics of the robot.In the path planning,the RRT* algorithm based on parent node reselection and rewiring strategy is used to obtain the optimized feasible paths more efficiently;meanwhile,the ellipse sampling mechanism is added to reduce the search range,which further shortens the path planning time and improves the overall efficiency;in the trajectory planning,the third and fifth polynomial In the trajectory planning,the seven-segment S-curve velocity planning scheme with smoother velocity and less damage to the robot is selected based on the comparison of the effects of the three and five times polynomial planning methods.Then,the calibration experiment of RGB-D camera is completed to obtain the camera-related parameters.Finally,two target detection and localization schemes are introduced: AKAZE feature point matching based on 2D images and Line Mod template matching method based on 3D point clouds.The advantages and disadvantages of the two schemes are analyzed through theory and experiment,and the correct identification and accurate grasping of rectangular boxes and badminton are completed using scheme two based on the ROS system. |
