Robotic arm grasping through 3D point clouds recognition

Abstract

The combination of 2D cameras and robots usually can no longer meet manufacturing production requirements. With the emergence of cheap 3D cameras, robot research based on 3D vision has become mainstream. In this paper, the Kinect camera is combined with the Fanuc manipulator to build an intelligent robot grasping system. First, we have proposed a new pentagonal positioning method, which can reduce errors in position conversion. Next, we designed our point cloud models for the model-based point clouds matching method. In the pose estimation process, we used a voxel grid to speed up the calculation, established a hash table that stores point pair features, and used Hough voting and pose Clustering to perform point cloud matching and output poses. Finally, we conducted several grasping experiments, and the experimental results met the requirements of grasping accuracy in our system.