This research introduces a novel Neural Network (NN) based approach to hand-eye calibration for Position-Based Visual Servoing (PBVS) in robotic manufacturing. The core of the work lies in utilizing a MultiLayer Perceptron NN to approximate the complex, nonlinear coordinate transformations required in visual servoing, specifically mapping image coordinates to real-world coordinates. This NN-based method offers the significant advantage of solving the hand-eye calibration problem without needing to explicitly estimate the hand-eye transformation, while simultaneously enhancing object tracking accuracy. The effectiveness of this proposed calibration method has been demonstrated through experimental results on an industrial manufacturing robot, showing superior performance compared to traditional methods for 2D object tracking.

The research presents a novel NN-based approach for hand-eye calibration in visual servoing, aiming to improve accuracy and simplify the calibration process in robotic manufacturing. The methodology focuses on approximating coordinate transformations using a MultiLayer Perceptron, which bypasses the need for direct estimation of the hand-eye transformation. Experimental results on an industrial robot indicate that this method outperforms existing techniques for 2D object tracking. The work contributes to the field by offering a more efficient and accurate calibration solution for precise robotic positioning.