Trajectory tracking control of unmanned mining truck considering actuator delays

Unmanned mining trucks operating in open-pit mines have substantial actuator delays, resulting in low trajectory tracking accuracy and poor operational stability. To address this problem, a trajectory tracking control strategy that considers actuator delays was proposed for unmanned mining trucks. First, a composite model comprising a first-order inertia element and a pure time-delay element was used to accurately model the delay characteristics of the lateral and longitudinal actuators. Based on this model, a longitudinal dynamic model and a lateral trajectory-tracking kinematic error model were established. Then, a decoupled lateral-longitudinal control architecture was adopted, and optimization problems were formulated for longitudinal Model Predictive Control (MPC) and lateral Nonlinear Model Predictive Control (NMPC), respectively. Tracking accuracy, operational economy, and driving safety were incorporated into the objective functions, while actuator performance limits were included in the constraints to ensure feasible control inputs and system stability. Finally, a bidirectional vehicle-state interaction mechanism over the prediction horizon was introduced into the decoupled architecture to coordinate the lateral and longitudinal controllers. Using actual road scenarios from an open-pit coal mine, Hardware-in-the-Loop (HIL) simulations and real-vehicle tests were conducted with a mining dump truck (model 930E), and the proposed controller was compared with an in-service vehicle controller already deployed at the mine. Results showed that the proposed controller outperformed the comparison scheme in tracking accuracy and control smoothness. Under simulation conditions, peak lateral, heading-angle, and velocity tracking errors were reduced by 65.5%, 77.1%, and 71.4%, respectively. In the real-vehicle tests, the corresponding peak errors were reduced by 61.3%, 57.1%, and 35.1%, respectively, and operational efficiency was improved by 9.6%. The proposed controller required an average of approximately 1.7 ms per computation step. This computation time meets the real-time control requirements of autonomous driving systems and further demonstrates the engineering feasibility of the proposed controller.