Crawling robot manipulator tracking based on gaussian mixture model of machine vision

Abstract

In the grasping process of the robot, the pose of robot should adapt to the change of the object's pose. In order to avoid the cumbersome calibration and difficulty of inversion in the existing vision-based robotic arm-grasping methods, a robotic arm grasping and tracking method based on Gaussian mixture of machine vision model (MGV) are proposed. This method uses Gaussian mixture model of machine vision to build the mapping relationship between the observed variables of the object and the robot joint variables. In the learning phase, the Gaussian mixture model of machine vision is used to directly construct the mapping from the pose of the target object to the joint angle of the manipulator. In the grasping stage, the pose of the target object is acquired through the camera, and the generation probability of the pose under each Gaussian component is calculated, respectively, and the Gaussian process regression corresponding to the Gaussian component with the largest posterior probability is selected to calculate the corresponding manipulator joint angle. The experimental results show that using Gaussian mixture machine vision model can make the robot manipulator grasp and track better than using single Gaussian process model.