A Novel DNN-Based Semi-Parametric Calibration Method for Parallel Robots Considering Non-Kinematic Parameters

Abstract

The pointing accuracy of pose adjusting parallel robots (PAPRs) is critical for the imaging quality of Cherenkov telescopes, making kinematic calibration crucial for improvement. However, non-geometric error sources like elastic deformation and joint clearances create an inevitable difference between theoretical and actual kinematic models, restricting accuracy. While various calibration methods exist, models considering non-kinematic nonlinear parameters are often highly complex and involve stringent measurement requirements. To achieve efficient, stable, and high-precision calibration, this paper proposes a novel deep neural network (DNN)-based semi-parametric model calibration method, combined with a Light-Spot-Mapping Kinematic Model (LKM). This approach enables high-precision matching between the theoretical and actual robot models through the semi-parametric framework, enhancing calibration performance. It also utilizes a high-precision camera as the measurement device, effectively reducing calibration time and labor costs. An experiment on a PS+PSS+PSR-type PAPR demonstrated that the pointing accuracy was improved by 47.7% after geometric parameter calibration, and by 81.2% (relative to the uncalibrated state) after non-geometric parameter calibration, achieving a final absolute accuracy of 156 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">${\mu rad}$</tex-math></inline-formula>. The proposed method is effective for improving the accuracy of PAPRs while accounting for the influence of nonlinear parameters.