Abstract: The pure pursuit model has become one of the core algorithms for path tracking of low-speed autonomous vehicles by virtue of its simplicity of structure, strong real-time performance, and engineering ease of implementation, and has been effectively validated in Ackermann steering vehicles. Different from the traditional rigid body, the articulated vehicle consists of the front and rear bodies connected by an articulation mechanism, and when choosing the front axle center or the rear axle center as the reference point (PPF or PPR), there are differences in the steering control volume generation mechanism, which directly affects the path tracking effect. The steering target computational model of a folded-waist steering articulated vehicle is constructed based on the pure pursuit algorithm, focusing on the mechanism of the influence of the reference point selection on the control accuracy, and at the same time, the load distribution is introduced to quantify the sensitivity of the reference point selection. A 3D model of articulated mining vehicle is built by Gazebo platform, and a LiDAR SLAM system is integrated within the ROS framework to plan the U-shaped test path and optimise the forward-looking distance parameters. The tracking performance of PPF and PPR is compared under no-load, half-load and full-load conditions, respectively. The experiments show that in U-turn tracking, the maximum lateral error and maximum heading error generated by PPF are smaller than those of PPR under all three loads, and compared with PPR, the maximum lateral error of PPF is reduced by 25.32%‒25.78% and the maximum heading error is reduced by 31.42% when no-load, and the maximum lateral error of PPF is reduced by 24.59%‒24.66% when half-load, and the maximum heading error is reduced by 31.06%‒31.37%; the maximum lateral error of PPF at full load is reduced by 22.37% and the maximum heading error is reduced by 30%‒32%. Meanwhile, as the loading mass of the front body increases, the tracking errors of PPF and PPR show a decreasing trend, and the maximum lateral error decreases by 3.85%‒8.07%, and the maximum heading error decreases by 2.77%‒5.55%. However, the degree of error reduction resulting from the increase in loading mass is significantly lower than the error variation caused by the change in reference point position parameters. This indicates that, compared with the load distribution state, path - tracking errors are more sensitive to reference point position parameters.