Hybrid Mapping Polynomial Neural Network Based on PSO Optimization for Robot Calibration

The absolute positioning accuracy of industrial robotic arms is crucial for advancing manufacturing applications such as automated assembly. This study presents a Hybrid Mapping Polynomial Neural Network based on Particle Swarm Optimization (PSO-HMPNN) for industrial robot calibration. The proposed approach integrates independent mapping neural networks with polynomial residual interaction neural modules to enhance error compensation capabilities, while employing the PSO algorithm to optimize neural network weights and avoid local optima. Experimental validation was performed on two distinct datasets (Dataset ABB and Dataset HSR) using ABB IRB120 and HSR JR680 robot, with performance compared against eight state-of-the-art calibration algorithms (M2-M9). Results demonstrate that PSO-HMPNN consistently achieved the lowest error metrics across both datasets. On the primary dataset (Dataset ABB), it reached a Mean Absolute Error (MAE) of 0.36mm, a Root Mean Square Error (RMSE) of 0.42mm, and a Maximum Error (MAX) of 0.92mm. This represents improvements of 14.3%, 19.2%, and 25.2% in MAE, RMSE, and MAX, respectively, over the best-performing baseline algorithm (M9). Similar high performance was also achieved on Dataset HSR, confirming the method’s robustness. Furthermore, the method shows excellent scalability, with an estimated computation time of 25.0s for a larger 500-sample dataset, achieving an optimal balance between accuracy and computational efficiency. The optimized Denavit-Hartenberg (D-H) parameters obtained after calibration meet industrial requirements, confirming the method’s superiority and applicability in robot calibration tasks.