MPC using an on-line TS fuzzy learning approach with application to autonomous driving

The control of complex nonlinear systems (such as autonomous vehicles) usually requires models which might be unavailable or inaccurate. In this paper, a novel data-driven Model Predictive Control (MPC) framework is proposed based on a data-driven approach to learn Takagi–Sugeno (TS) fuzzy models for nonlinear systems. To address the data TS modeling, we use the Evolving TS Fuzzy Ellipsoidal Information Granules (TS-EEFIG) approach to obtain a polytopic representation as well as a set of membership functions that allows to use efficient linear control tools to handle complex nonlinear systems. In particular, the formulated approach is applied for the autonomous driving control problem of a racing vehicle. The proposed control uses references provided by an external trajectory planner offering a high driving performance in racing mode. The control-estimation scheme is validated in a simulated racing environment based on a high fidelity vehicle model of a 1/10 scale RC car to show the potential of the proposed approach.
Peer Reviewed
Postprint (author's final draft)