Multitime-Scale Model Predictive Control Method for Robot Grasping Based on Visual Servoing

Authors:
Publisher website
Google Scholar
Abstract
Concerning the problem of inaccurate target localization and trajectory tracking in robot grasping. This paper presents a multi-time-scale model predictive control method for robot grasping based on visual servoing. First, the depth camera captures the depth information and 2-D images of the target and fuses them to generate a point cloud. Then, the mathematical model of position with depth variation is established by using the characterization of the point cloud reflecting the target position information. Next, considering the multi-time-scale characterization of the trajectory, it is decomposed into several subsequences. The support vector regression (SVR) prediction model is also established to predict future subsequences. Tracking of the target trajectory is realized by applying control to the robot. Finally, some experimental results based on actual industry data show that the method controls the robot to perform grasping tasks accurately in the complex industry field.
Funding Information
  • National Natural Science Foundation of China (62373336)
  • 111 Project, China, (B17040)

This publication has 34 references indexed in Scilit: