Dual-Loop neural prediction– based trajectory tracking control for robotic arms
Abstract
With the rapid growth of intelligent manufacturing, robotic arm trajectory tracking control faces challenges of high precision and fast response. Existing control methods suffer from lagging parameter convergence, loss of prediction accuracy at single time scales, and other issues when dealing with joint coupling nonlinearity, load mutations, and impact disturbances, resulting in large tracking errors and insufficient real-time response performance. To this end, a robotic arm trajectory tracking control model was proposed based on dual-loop neural prediction. First, a collaborative prediction mechanism of inner and outer dual-loop neural networks is constructed. The inner loop network captures the system's nonlinear dynamics in real time and performs short-term trajectory prediction. In contrast, the outer loop network performs long-term trend prediction and systematic error compensation based on historical data and environmental feedback. Second, an online learning strategy based on prediction error gradients is designed to accelerate parameter convergence and enhance the system's adaptability to external disturbances. Finally, a fusion control framework combining neural prediction and traditional control is constructed, achieving a balance between computational efficiency and control accuracy through the complementary and coordinated functions of global optimization and local stability guarantee. Experimental results show that our model achieves 0.82mm tracking accuracy on the LASA dataset and 0.91mm on the RoboSet dataset, with response times of 12.3ms and 14.2ms, respectively, significantly outperforming existing methods. The disturbance rejection capability reaches 42.7% and 39.8%, demonstrating excellent robustness, while maintaining computational efficiency improvements of 16.8% and 14.3% respectively.