A Vision-Based Error Compensation Method for Accurate Robotic Positioning in Automated Marking

Abstract

Accurate robotic positioning is essential in automated marking because residual pose error directly shifts the permanent mark. This study presents a one-step vision-based residual error compensation method that measures the displacement between the target and the marking-tool position after the initial robot motion, transforms that displacement into the robot base frame, and applies it as a translational Tool Center Point (TCP) correction. The platform combines an AR4-MK3 six-degree-of-freedom manipulator, a fixed Intel RealSense D455 RGB-D camera, and a 50 W electromagnetic marking device. At five spatially distinct target points, the mean Euclidean positioning error decreased from 7.40 ± 1.35 mm to 2.14 ± 0.36 mm. The mean paired decrease was 5.26 mm (95% confidence interval: 4.04-6.48 mm), and all five points improved by 70.7%-71.4%. An exploratory point-level paired test gave t(4) = 11.92 and p < 0.001. The result supports the feasibility of inserting a local visual correction immediately before marking without changing the nominal robot kinematic model. The evaluation is a five-point pilot rather than a repeatability or full workspace study; raw repeated-trial, orientation-error, timing, robustness, and quantitative mark-quality measurements were not available. The conclusions are therefore limited to translational positioning under the tested camera and workcell configuration.