High-Precision Monocular Vision Guided Robotic Assembly Based on Local Pose Invariance
Xu Rui, Xingwei Zhao, Feng Liu, Bo Tao
Abstract
Industrial robots provide a new production mode for automatic assembly. However, the motion accuracy of robots is hard to meet high-precision assembly requirements in many current scenarios. To address this issue, this paper proposes a robotics assembly strategy based on local pose invariance, achieving high-precision assembly guided by a monocular camera. This assembly strategy combines monocular measurements with local reference poses, providing a new assembly pose solution chain that effectively mitigates the accumulation of errors caused by the lengthy coordinate chain propagation. Next, a monocular visual measurement method based on 3D-2D matching is introduced, which improves the traditional matching model and enhances the efficiency and accuracy of the measurement process by directly mapping 2D pixel coordinates to 3D coordinates. This measurement method is universal and applicable to both calibration and actual assembly. Experimental results demonstrate that the proposed assembly method is suitable for both hole-to-hole and peg-in-hole tasks and can adapt to assembling different components, realizing final error better than 0.26 mm within a range of approximately 4 m.