With the continuous advancement of construction technologies toward intelligence, robotic rollers equipped with intelligent compaction techniques have gradually become a research focus in the field of infrastructure engineering. During operation, the path-tracking accuracy of robotic rollers plays a vital role in ensuring compaction quality and improving efficiency. This study addresses the path-tracking control problem of robotic rollers operating on unstructured road surfaces. A motion model and a multi-body dynamic model are developed, based on which a Model Predictive Control (MPC) strategy is introduced. To validate the effectiveness and robustness of the proposed control method, a co-simulation platform integrating RecurDyn-MATLAB-EDEM is established to simulate the robotic roller’s behavior under complex conditions. Comparative analyses are conducted between the MPC approach and traditional methods, including Proportional-Integral-Derivative (PID) control and Pure Pursuit Control (PPC). Simulation results demonstrate that under initial tracking deviations, the MPC algorithm can reduce and stabilize errors within ±5 cm in 15 s. Compared to PID and PPC methods, the MPC strategy reduces both the maximum and average tracking errors by over 60%, highlighting its superior control stability and adaptability to varying environments. These findings provide theoretical and methodological support for intelligent compaction in complex construction scenarios.