Research On Obstacle Avoidance Motion Planning Of Manipulator Based On Visual Perception

With the strategy of "industry 4.0" and "made in China 2025" put forward one after another,robots have once again entered the public field of vision.As a large class of robots,manipulator is often active in the production line of many factories because of its tireless nature of work,as well as incomparable accuracy and stability,but it is usually completed by manual teaching or some pre-defined programming,so that they can replace manual to complete some repetitive work.But the flexibility and intelligence of the manipulator is still a critical problem,so it has been the direction of researchers to make it have higher flexibility and autonomy.In view of the above problems,the depth camera can not only collect the image information in the environment,but also get the depth information of the corresponding points in the image,so this paper uses the depth camera as the eye of the manipulator to collect the environmental data.Firstly,through the visual perception information processing module,the Kinect V2 depth camera is used to collect the point cloud data of the manipulator workspace.After data processing,the environment model of the manipulator workspace is obtained.Then,the multi-level and multi type bounding box is used to envelope the obstacles in the environment through the multi type target collision detection module,and a variety of constraints are attached to the manipulator.Finally,motion planning is carried out based on the improved RRT algorithm,and the effectiveness of the algorithm is verified in the experimental environment.The main contents of this paper are as follows:In this paper,the improved Denavit-Hartenberg(DH)parameters are used to establish the coordinate system of AuBo-i5 manipulator used in the laboratory,and the link parameters and coordinate system transformation matrix of the manipulator are given.On this basis,the forward and inverse kinematics analysis is completed,and the correctness of the established kinematics model is verified by simulation experiments.In this paper,a visual perception information processing method is proposed,which collects the working environment information of the manipulator through the visual sensor.Firstly,aiming at the distortion and error that may be caused by light passing through multiple lenses in the process of camera imaging,the camera is calibrated with internal parameters to reduce the data acquisition error.The camera is fixed outside the manipulator to complete the hand eye calibration,which lays the foundation of coordinate transformation between the manipulator and the camera.Then,the point cloud data collected by the camera is processed.Compressed sampling space,radius filtering algorithm and improved voxel filtering algorithm are used to focus the attention around the manipulator to further optimize the collected information,and octree is used to optimize the storage space of the point cloud.Then a multi type collision detection method is proposed,which uses multi type bounding box to envelope the working environment of the manipulator.Because the simplicity of bounding box and its compactness are often contradictory,and combined with the fact that there may be many shapes of objects in the actual environment,the method uses sphere,cuboid and cylinder to envelope the objects in the working environment of the manipulator,and then uses a layer of sphere bounding box in the outer layer,and adds a variety of constraints,so as to achieve the purpose of fast collision detection,It lays a foundation for the interaction between the manipulator and objects in the workspace.Finally,based on the workspace information of the manipulator,combined with the improved RRT algorithm,the motion planning of the manipulator is carried out.Methods the extended point selection strategy is adopted to make the algorithm expand to the target point quickly and purposefully,and reduce the number of iterations.The adaptive step strategy is adopted to make use of the environment information obtained from the current exploration in the process of moving,so that the tree can cover less obstructed spatial areas faster,and the local escape mechanism is introduced to avoid falling into the local minimum.Finally,the algorithm combines the greedy idea of Dijkstra algorithm to further optimize the path.Then,a manipulator planning strategy is proposed for the obtained path,which greatly reduces the motion cost of the manipulator.