Research On Vision-based Object Grasping Technology Of Mobile Manipulator

The mobile manipulator has both the autonomous mobility of a mobile robot and the operation capability of the manipulator.It has a wide range of applications.It can alleviate the labor shortage caused by the aging of the population in the industrial field and improve the lives of the people in the service industry.quality.However,in the unstructured indoor environment,the target grasping technology of mobile robotic arms is far from mature.Based on the indoor environment,this paper constructs a two-wheel differential mobile manipulator system for the application requirements of the mobile manipulator to autonomously grab the specified target object,and conducts research on the main technologies involved,including the path planning of the mobile manipulator,Target object recognition and positioning,manipulator motion planning.In order to realize the path planning of the mobile manipulator in the scene map,the principles of Dijkstra and A* global path planning algorithms is analyzed.Through simulation experiments,it is found that the A* algorithm is more efficient than the Dijkstra algorithm path planning,and the A* algorithm is selected carry out global path planning.Aiming at the obstacle avoidance problem of the mobile manipulator in the movement process,the dynamic window approach(DWA)is used for local path planning.The twowheel differential motion chassis model is analyzed,the planning principle of the DWA algorithm is studied,and the method of controlling variables is adopted to obtain the optimal parameters of the DWA algorithm evaluation function through simulation experiments.In order to obtain the pose information of the target object,this paper carried out the research of object recognition and positioning algorithm based on RGB-D camera.Aiming at the large computational complexity of the point cloud feature extraction method,this paper proposes an object detection algorithm combining 2D image and 3D point cloud.First,extract the SURF features of the two-dimensional image for template matching,and use the fast library for approximate nearest neighbors(FLANN)to improve the matching efficiency.Secondly,the homography projection transformation is used to find the position of the target object in the scene image as the region of interest.Finally,the ICP algorithm is used to register the point cloud of the scene region of interest with the template point cloud to realize the object pose estimation.Based on the target pose information obtained by the vision system,this paper studies the joint space trajectory planning algorithm of the manipulator.the four-degree-of-freedom manipulator is modeled by the improved D-H method,and the problem of forward and inverse kinematics is analyzed.in order to solve the jitter problem in the traditional quintic spline curve trajectory planning,this paper proposes to use the seventh-order polynomial trajectory planning algorithm for target grabbing trajectory planning.The simulation results show that the motion trajectory of the manipulator is smooth,there is no jitter in the motion process,and the algorithm stability is good.Based on the above research work,the functional configuration of the mobile manipulator is completed in the ROS system,and the related algorithms studied are introduced into the control system,and experiments of autonomous positioning and navigation and target recognition and grasping are carried out.The experimental results show that in a unstructured indoor scene,the mobile manipulator can move to the target position according to the planned path,accurately recognize the target object and complete the grasping task,and meet the design requirements.