Abstract: Autonomous vehicles frequently negotiate sharp turns, U-shaped curves, and continuous high-curvature paths in applications such as mining areas, ports, and industrial parks. Under such conditions, path-tracking control systems face the fundamental challenge of simultaneously ensuring tracking accuracy, real-time performance, and driving efficiency, particularly when the vehicle steering capability and computational resources are limited. Existing path-tracking methods, particularly those based on conservative constant-speed strategies, often sacrifice driving efficiency to maintain accuracy, while methods that rely on nonlinear model predictive control (NMPC) often incur excessive computational burdens in real-time applications. To address this problem, this paper proposes an active speed adjustment path-tracking control method based on time-varying local (TVL) linear model predictive control (LMPC), specifically designed for large-curvature path scenarios. The proposed method introduces a time-varying local coordinate system in which the vehicle dynamics model is reformulated and linearized under the small-angle assumption. This transformation enables the construction of a linear predictive model that remains valid over the prediction horizon, even when the vehicle traverses paths such as right-angle turns and U-shaped curves. On this basis, a multireference point strategy is incorporated into the LMPC framework, allowing the controller to explicitly account for the geometric characteristics of high-curvature paths to improve the tracking accuracy compared to conventional single-reference-point LMPC approaches. To resolve the inherent conflict among tracking accuracy, real-time performance, and driving efficiency, a rule-based active speed adjustment strategy was introduced. The strategy dynamically adjusts the vehicle speed based on the current vehicle state, path curvature, and adhesion limits to ensure that the tracking accuracy and driving efficiency are maintained. The effectiveness of the proposed TVL–LMPC-based active speed adjustment method was validated through comprehensive MATLAB–CarSim co-simulations and hardware-in-the-loop (HIL) testing. Co-simulation results demonstrate that the proposed method can stably complete path-tracking tasks at both low and high speeds on large-curvature paths, with a maximum displacement error of 0.1271 m. In contrast, the constant-speed TVL–LMPC, active speed-adjusted global LMPC, and active speed-adjusted NMPC exhibited a divergence in error when negotiating sharp turns, highlighting the superior suitability of the proposed approach for complex path scenarios. HIL tests were conducted to evaluate the performance of the proposed method under conditions similar to those in practical deployment. To assess the robustness to sensing uncertainty, random positioning errors of ±1 cm were introduced, consistent with typical positioning system performance. Under these disturbance conditions, the maximum displacement error remained below 0.1336 m, indicating that the proposed control system maintained a stable and accurate tracking performance despite localization disturbances. In terms of real-time performance, under identical simulation conditions, the proposed method reduced the maximum and average computation times by at least 27.95% and 25.61%, respectively, compared with the active speed-adjusted NMPC, confirming its suitability for real-time implementation. Overall, the proposed TVL–LMPC-based active speed adjustment path-tracking control method effectively achieves precise, real-time, and efficient path tracking on large-curvature paths. The method is particularly suited for autonomous vehicle applications in mining areas, ports, and industrial parks, where complex path geometries, real-time constraints, and operational efficiency are critical.