Hybrid Model-Predictive Control for Autonomous Ground Vehicle Control

Charles Jawny (Nova Southeastern University), Yesenia Velasco (North Carolina Central University)

Model predictive control (MPC) has gained recent interest in dynamic systems due to its ability to optimize solutions in real time while in view of constraints and future events, i.e. obstacles or the environment. In this paper uncontrollable divergence is used as a main factor for our switching logic of a hybrid MPC to alternate between several reference models of an autonomous vehicle with a satisfactory level of model mismatch. The inclusion of multiple reference models allows for more effective maneuvering of an autonomous vehicle due to the fact that using a single reference model is not optimal for all driving scenarios. The effectiveness of our solution is demonstrated in a simulation of an autonomous vehicle using the hybrid MPC to maneuver successfully through an obstacle course at a lower cost than would be if using a single reference model. Our final results are quantified in a real autonomous vehicle. This work could help improve the robustness of autonomous vehicular development.