Robot Disordered Picking Using Point Pair Feature Algorithm

Picking target objects is one of the most common operations of robots, and how to achieve the disorderly picking of multiple workpieces during the product production process is an important area of research for robot applications. The article firstly briefly analyzes point cloud registration of traditional machine vision and the development of robot picking algorithms based on deep learning; Taking picking specified location of the target object into consideration, we choose PPF algorithm to estimate initial pose of workpiece. Secondly, a framework for robot disorderly picking process based on multiple machine vision algorithms was established, and the principle of PPF algorithm and improved binocular calibration method were introduced in detail. Finally, a real-world scene was built for testing of the framework. ROS interface, point cloud imaging and accessibility for several stereo cameras were tested in ROS environment. Hand-eye calibration was completed using an ArUco marker plate and easy_handeye package. Then workpiece model point cloud and segmented scene point cloud were registered and the robot’s disordered picking operation tests were achieved.