Research Information System
IEEE Access, vol.14, pp.24330-24338, 2026 (SCI-Expanded, Scopus)
This paper introduces a Neural Network Augmented Koopman Explicit Model Following Control (NNKEMFC) framework for robust trajectory tracking of robotic manipulators with nonlinear effects and uncertain dynamics. Unlike existing Koopman operator based or purely neural control approaches, the proposed method explicitly separates data-driven modeling and online residual learning within a structured model following control architecture. An offline-identified extended-state Koopman operator is implemented into an Explicit Model Following Control (EMFC) framework to generate a structured nominal feedforward torque while preserving interpretability and control transparency. To compensate for modeling errors, friction effects, external disturbances, and parametric uncertainties, an online neural residual compensator is incorporated to estimate unmodeled dynamics in real time. The final control input consists of the Koopman-based feedforward torque, a filtered tracking-error feedback term, neural compensation, and an additional robustifying signal to guarantee stability enhancement. The proposed framework is evaluated on a multi-degree-of-freedom robotic manipulator under friction, external loads, and parameter uncertainties. Simulation results indicate that the NNKEMFC consistently outperforms classical computed torque control and KEMFC, resulting in significant reductions in joint-space RMS tracking errors in the existence of nonlinearities. These results affirm that the explicit integration of a structured Koopman-based model followed by targeted neural residual learning offers a scalable and robust solution for achieving high-precision robotic control.