Practical Insights on Data-Based Robot Control: A Comparative Analysis of Data-Enabled Predictive Control and Model-Based Predictive Control

Abstract When partial knowledge of a system is already available, is it reasonable to disregard this information in favor of a purely data-driven control approach? In this study, Data-Enabled Predictive Control (DeePC) is implemented on a custom-built omnidirectional mobile robot and a comparative evaluation against Model Predictive Control (MPC) is conducted seeking to address this question. To ensure a fair comparison, a linear model used for MPC is identified from the same dataset employed by DeePC. Additionally, a supplementary comparison using a nonlinear nominal model is provided. All results are derived from hardware experiments, which are essential for assessing the real-world applicability of DeePC, particularly in robotics, where system dynamics is often nonlinear and data is constantly interfered with by noise. The findings offer practical insights into the challenges and strengths of DeePC, which has been claimed to be robust to nonlinearity in the literature, aiding practitioners in selecting the most suitable predictive control strategy for their specific purposes.