Pose accuracy compensation of mobile industry robot with binocular vision measurement and deep belief network

Abstract It is very important to improve the absolute pose accuracy of the robot, in the automatic drilling and riveting system of mobile industrial robot. This paper presents a practical scheme to compensate for the pose accuracy of mobile industrial robot. Firstly, the binocular vision measurement of the robot pose was building, and the pose error samples were collected under the massive configuration of the robot. Secondly, the mapping model of the robot's theoretical pose and actual pose errors is established based on the deep belief network, and the pose error estimation is realized. Finally, the developed scheme has been verified on a mobile industrial robot KUKA KR500-3 of robot automatic drilling and riveting system. The proposed scheme took into account the position error and direction error of the robot into account, does not depend on the specific kinematics model of the robot, and without retrofitting with high-end encoder. The verification results showed that the maximum absolute position error and orientation error of the robot was reduced by 80.84 % from 1.524 mm to 0.292 mm and reduced by 57.29 % from 0.096 deg to 0.041 deg respectively, which can meet the requirements of the robotic automatic drilling and riveting system.