Improved Integral Sliding Mode Control for AUV Trajectory Tracking Based on Deep Reinforcement Learning

Trajectory tracking control of autonomous underwater vehicles (AUVs) faces challenges in complex nearshore environments due to model uncertainties and external environmental disturbances. Traditional control methods often rely on expert knowledge and manual parameter tuning, which limit the adaptability of AUVs to structural variations and changing operating conditions. Moreover, inappropriate parameter selection in conventional sliding mode control may induce high-frequency chattering, degrading control accuracy and operational efficiency. To address these issues, this paper proposes an improved integral sliding mode control (IISMC) strategy integrated with deep reinforcement learning (DRL). In the proposed framework, DRL is employed to adaptively tune key controller parameters, including the sliding surface coefficients and reaching law gains, while preserving the analytical structure of the IISMC scheme. This adaptive tuning mechanism effectively suppresses chattering and enhances robustness against uncertainties and disturbances. Numerical simulation results demonstrate that the proposed DRL-assisted IISMC method achieves improved disturbance rejection capability, higher trajectory tracking accuracy, and smoother control performance compared with conventional sliding mode control (SMC) approaches under identical operating conditions.