Comparative Analysis of Kinematic and Identified MIMO Models in Model Predictive Control for Mobile Robot Trajectory Tracking

Trajectory tracking remains one of the main challenges in mobile robotics, particularly when robots operate under real-world disturbances and modeling uncertainties. Model Predictive Control (MPC) has become one of the most effective solutions for this problem because of its ability to optimize future control actions while explicitly handling system constraints. However, despite the variety of predictive models reported in the literature, there is still limited experimental evidence regarding how predictive-model selection influences the closed-loop behavior of mobile robots. This work presents an experimental comparison between two MPC implementations: a Kinematic MPC and a multiple-input multiple-output MPC (MIMO-MPC) based on an experimentally identified state-space model. Both controllers were implemented on the same four-wheel differential-drive mobile robot running Robot Operating System 2 (ROS 2) and were configured with identical prediction horizons, weighting matrices, constraints, reference trajectories, and disturbance sequences, enabling a direct experimental comparison of the influence of predictive-model selection on closed-loop performance. Performance was assessed through trajectory-tracking accuracy, disturbance recovery, and control smoothness metrics under both nominal and externally perturbed operating conditions. The experimental evaluation was performed on real hardware at 50 Hz using a closed-loop trajectory of approximately 20 m. Under nominal conditions, the kinematic MPC achieved a lower lateral root mean square error (RMSE) (0.1031 m versus 0.1187 m) and maintained 93.23% of the trajectory within the ±0.20 m tolerance band. Under external perturbations, however, the MIMO-MPC reduced the heading RMSE from 0.7617 rad to 0.6503 rad, shortened the recovery time after the two largest disturbances by up to 35.6%, and decreased the angular-velocity rate root mean square (RateRMS) and jerk root mean square (JerkRMS) by approximately 24.7% and 24.8%, respectively. These results demonstrate that predictive-model selection has a significant influence on the transient behavior of MPC. While a conventional kinematic model provides excellent nominal tracking performance, an experimentally identified MIMO predictive model offers faster recovery and smoother control under disturbed conditions. The proposed experimental methodology provides practical evidence that can assist the selection of predictive models for MPC-based mobile robots according to the expected operating conditions.