Path tracking and obstacle avoidance algorithm for underground load–haul–dump machines based on ILC-TEB

Underground load–haul–dump machines are affected by high-frequency disturbances caused by roadway surface irregularities, sensor noise, and motor vibration, making it difficult for conventional path-tracking algorithms to achieve both tracking accuracy and obstacle avoidance. To address these problems, a path-tracking and obstacle-avoidance algorithm combining Iterative Learning Control (ILC) and Timed Elastic Band (TEB) was proposed. To address iteration failure in conventional time-domain ILC, path arc length was used as an independent variable for iterative optimization based on vehicle trajectories. To meet tracking requirements on sections with different curvatures, which are difficult to satisfy using fixed-gain Proportional–Derivative (PD) ILC, an error-based adaptive gain adjustment mechanism was adopted to dynamically adjust proportional and derivative gains according to roadway conditions. A second-order Butterworth low-pass filter was introduced as a Q-filter to suppress high-frequency disturbances and enhance robustness. To prevent erroneous ILC corrections caused by obstacle-avoidance detours, a spatial correlation decay function was incorporated into the ILC learning law to adaptively adjust the contribution of historical errors to current corrections, maintaining convergence stability while allowing flexible obstacle avoidance. Simulation results showed that ILC-TEB achieved lower mean lateral and heading-angle errors than PID, Model Predictive Control (MPC), and TEB under both straight and curved roadway conditions. PID failed to avoid obstacles under straight roadway conditions, and both PID and MPC failed under curved roadway conditions, whereas TEB and ILC-TEB successfully avoided obstacles under both conditions. In real-vehicle tests involving three typical obstacles—spoil heaps, vehicles, and ore piles—ILC-TEB reduced maximum distances from obstacles by 20.1%, 15.0%, and 22.3%, respectively, compared with TEB, reducing unnecessary detour space. Mean lateral and heading-angle errors decreased by 22.02% and 18.80%, respectively. These results demonstrate improved path-tracking accuracy.