Predictive controller design for image-based visual servoing in robot manipulator using event-triggered MPC and physics informed neural network

This paper proposes a novel physics-informed neural network (PINN)-based event-triggered model predictive control (EMPC) framework for trajectory planning in image-based visual servoing (IBVS). The main novelty lies in the design of a PINN that directly predicts feature-point updates from visual feedback, eliminating the explicit computation of image and velocity Jacobians while incorporating the underlying visual servoing physics. The proposed EMPC integrates the trained PINN model into the optimal control problem (OCP) to generate manipulator trajectories from camera feedback. The PINN-based dynamics prediction and event-triggered optimisation strategy reduce the task execution time compared to conventional schemes. Experimental validation on a 5-DOF ViperX 300 manipulator demonstrates the effectiveness of the proposed algorithm. The developed PINN achieves a 23.57% improvement in Jacobian approximation accuracy over conventional neural networks, while the proposed PINN-based EMPC reduces computation time by 29.23% and tracking error by 5.7% compared with the analytical model-based MPC scheme.