Nonsingular fast terminal sliding mode control for robotic manipulators: a synergistic approach integrating RBFNN and AESO

This study proposes a composite control architecture combining radial basis function neural network (RBFNN) and adaptive gain extended state observer (AESO) to address trajectory tracking challenges in robotic manipulators under nonlinear dynamics, parametric uncertainties, and external disturbances. The framework integrates three core innovations: (1) a global nonsingular fast terminal sliding mode surface with finite-time convergence and singularity avoidance, which is augmented by an integral term for enhanced robustness and steady-state error suppression; (2) a dual compensation mechanism where RBFNN approximates unmodeled dynamics through nonlinear mapping, while AESO provides real-time estimation of lumped disturbances; (3) a stability guarantee derived through Lyapunov-based synthesis. Comparative simulations on a two-link manipulator demonstrate superior performance over conventional single compensation methods, with significant improvements in convergence speed, tracking precision, and torque smoothness. The results validate the efficacy of the framework in complex operational scenarios. However, although AESO can balance state estimation accuracy and noise immunity by dynamically adjusting bandwidth, noise amplification is inevitable when dealing with high-frequency noise due to inherent defects of gain adaptive mechanism.