Gradient Fuzzy Neural Dynamics-Based Hybrid Learning Control for Unknown Robot Manipulators Under Boundary Constraint

A precise understanding of the manipulator model is essential for effective tracking control in robotic applications. Once the kinematic model of the manipulator is unavailable, the application of model-based methods may result in the inability to successfully accomplish a desired task. To this end, a hybrid learning control framework is proposed to enhance model accuracy and trajectory tracking performance, which integrates gradient fuzzy neural dynamics (GFND) with a varying-parameter predefined-time convergence term. With the aid of GFND, the Jacobian matrix of the manipulator is accurately estimated, and fuzzy logic systems are adopted to adaptively tune the learning rate. The incorporation of a predefined-time framework ensures that robotic control tasks are completed within a specified duration, regardless of the initial system states. To ensure adherence to joint constraints in practical settings, a nonlinear mapping function is designed, which in turn enhances the manipulator’s ability to accomplish designated control objectives. The proposed algorithm is proven to converge through comprehensive theoretical analyses, while its feasibility and advantages are demonstrated through simulation and experiment on a manipulator.