Optimizing computational efficiency and precision of UR-5 robotic manipulator through MRSM-based reliability analysis

The increasing deployment of robotic manipulators in precision-critical applications demands accurate trajectory tracking and reliable performance under uncertainty. Traditional methods, such as Monte Carlo Simulation (MCS), provide robust reliability estimates but are computationally prohibitive for real-time use. Surrogate models, including Support Vector Regression (SVR) and Neural Networks (NN), mitigate this limitation but often lack uncertainty quantification and require extensive training data for convergence. Accurately predicting end-effector behavior across multi-dimensional output spaces remains a persistent challenge. To address these issues, this study proposes a Multi-Response Surrogate Model (MRSM) based on Multi-Output Gaussian Processes with adaptive sampling to estimate the time-dependent reliability of 6-DOF robotic manipulators. By capturing inter-output correlations and iteratively refining high-error regions, the MRSM significantly improves predictive accuracy while reducing computational cost. Validation on the UR5 manipulator demonstrates close agreement with MCS results, achieving a 98% reduction in computational effort and outperforming SVR and NN in both accuracy and refinement efficiency. A precision cutting task on fiberglass and carbon fiber composites further confirms the model’s practical applicability, maintaining a ± 0.25 mm tolerance under constrained dynamics. These results demonstrate the MRSM’s suitability for high-precision, uncertainty-aware manufacturing and advanced composite fabrication.