Reinforcement learning‐enhanced adaptive sliding mode controller for underground articulated vehicles trajectory tracking

Abstract

High‐precision trajectory tracking for underground articulated load‐haul‐dump (LHD) vehicles is significantly challenged by strong kinematic coupling and drastic payload variations. While sliding mode control (SMC) provides inherent robustness, its performance is often limited by fixed parameters that fail to adapt to varying path geometries. This paper proposes a reinforcement‐learning‐enhanced adaptive sliding mode control (RL‐SMC) framework, which utilizes a reinforcement learning agent for the online optimization of critical SMC parameters. A core innovation is the integration of a trajectory preview mechanism designed to enhance steering maneuverability and ensure operational safety within the narrow confines of underground drifts. Comprehensive numerical simulations and full‐scale field experiments demonstrate that the RL‐SMC framework outperforms both fixed‐parameter SMC and pure RL approaches, maintaining an 11.71 cm mean tracking error and a 1.83 ms solving time even on slippery road surfaces. These results confirm the framework's superior precision and real‐time efficiency, offering a robust and deployable solution for autonomous mining haulage.