Data-Driven Identification Based Model Predictive Control for a Ground Robot

Unmanned Ground Robots (UGVs) are commonly used test beds for the design and validation of Guidance, Navigation and Control algorithms. Thanks to the onboard sensors, it is possible to extract information about the system state, including position, velocity or orientation. Additionally, data can be stored and examined for various purposes, such as the creation of a model for the system. Predictions about the system state can be exploited by a model-based predictive controller to decide the best control action to be carried out at the current time. This paper proposes a Tracking Linear Quadratic Model Predictive Controller (LQMPC) for a UGV robot, together with a Kalman Filter (KF) estimator for state retrieval. The system model is identified with a mixed approach, combining a kinematic model with a data driven model whose data have been collected during experiments with the real platform. The effectiveness of the data driven approach for MPC is then proven for a velocity and orientation tracking application in a simulative scenario.