To address the problem of insufficient control accuracy and robustness of automated guided vehicles (AGVs) under nonlinear time-varying conditions and external disturbances, a data-driven adaptive discrete-time integral sliding mode control (ADISMC) method is proposed. This method integrates three core strategies: first, full-form dynamic linearization (FFDL) is employed to transform the unknown nonlinear system into a compact-form dynamic model relying solely on I/O data, thereby circumventing the dependence on mechanistic modeling in principle; second, a BP neural network observer (BPNNO) is constructed to perform online estimation and feedforward compensation of the lumped disturbance, which substantially reduces the switching gain while ensuring robustness, thus mitigating chattering at its source; third, an adaptive reaching law with the tracking error as the independent variable is designed, achieving fast convergence when far from the sliding surface and automatic gain contraction when approaching the sliding surface, thereby balancing response speed and control smoothness. Theoretical analysis rigorously proves the boundedness of the pseudo gradient and the finite-time convergence of the sliding mode motion. AGV kinematic simulations demonstrate that the proposed method significantly outperforms PID and standard MFAC in overall performance: the maximum heading tracking error under disturbance-free conditions is merely 0.3488 rad, and the mean square error under 0.3 rad white noise disturbance is only 14% of that of the PID scheme, validating the superior tracking accuracy and disturbance rejection robustness of the proposed method.
Xu et al. (Mon,) studied this question.