Intelligent Picking And Palletizing Planning Of Muti-dimensional Boxes Based On RGB-D

In the field of traditional robotic e-commerce logistics box sorting and palletizing,usually only a single task needs to be completed,either only the box sorting task or only the box palletizing task.It is rare to combine box sorting task and box palletizing task.But now more and more factories need to adopt flexible production mode.To deal with different sizes boxes,it needs to finish box palletizing and packing steps after box sorted.Therefore,there is a problem with the production mode of sorting or palletizing separately.Aiming at solving the above problems,this paper studies the related problems of the combination of robotic box sorting and palletizing in the e-commerce logistics scenes.The specific research content is as follows:(1)A sorting method of multi-size boxes based on RGB-D is proposed.Firstly,Azure Kinect is selected to obtain the RGB-D information of different boxes,and the camera is performed calibration work.Then,use the point cloud instance segmentation and labeling tool easylabel to construct the data set required for deep learning model training,and select the efficient and simple point cloud instance segmentation network 3D-Bo Net for box segmentation.Finally,refine mis-segmented box planes by means of merging nearest neighbor points in terms of different box planes concave package,and improve the accuracy of cabinet instance segmentation.(2)A sorting strategy based on multi-parameter fusion is proposed.After using the point cloud segmentation method based on deep learning to obtain the instances of different boxes,extract a grab box face element from each box,and obtain the grab point pose of the face element by analyzing its six-dimensional pose.Then,the size of the box grabbing panel and the angle between the panel normal and the Z axis of the camera coordinate system,the center position of the grabbing point,and the inverse kinematics solution of the grabbing point are combined to design a multi-parameter fusion scoring strategy.In the actual box sorting process,the box with high score is sorted first.(3)A palletizing planning method using both volume and area to optimize is proposed.After investigating the commonly used palletizing planning algorithms in the scenarios of this paper,and reproducing the volume-based multi-size box palletizing planning,a palletizing planning method using both volume and area to optimize is proposed.Finally,a new box size data set is designed,and the effectiveness of the algorithm proposed in this paper is verified on the original data set and the new data set.(4)The box sorting and palletizing planning system has been built.Integrate the box point cloud segmentation algorithm,the best point grab algorithm and the visual servo algorithm into the sorting system,use the RTDE(Real-Time Data Exchange)software package to control the robotic arm to complete the box sorting experiment,and obtain the sorting box size data.Then,the palletizing planning algorithm is integrated into the palletizing system,and the palletizing planning algorithm is used to obtain the palletizing sequence of each box based on the data obtained by the box sorting process.Finally,the UR10 e robot is used to complete the palletizing of the box palletizing experiment.The experiment verifies the feasibility of the integrated scheme of box sorting and palletizing in this paper.