Pushing the boundaries of nonlinear data-driven MPC

Autonomous systems, such as self-driving cars and drones, rely on advanced control algorithms to make decisions and navigate complex environments.

Model Predictive Control (MPC) is a popular control strategy for complex tasks due to its ability to handle constraints and optimize the control actions. However, traditional MPC methods are limited by the accuracy and availability of system models.

With the increasing availability of sensor data and advances in data-driven modeling techniques, data-driven MPC has emerged as a promising approach to cut the deployment time and to improve the performance and robustness of MPC.The overall goal of this proposal, to carried out by the PI and a PhD student over 4-years, is to push the boundaries of nonlinear data-driven MPC, by first relying on recent results on "Takens embedding", a tool that allows to reconstruct the state of a generic nonlinear dynamical system.

Then, we plan to develop and extend neural networks, known as echo state networks, to act as state estimators for data-driven MPC.

Finally, we will focus on how to design experiments to collect informative data for these state estimators to perform reliably.This proposal constitutes a novel framework for nonlinear data-driven MPC, which has so far been mostly limited to the sub-class of linear systems.

This, we expect, will open the doors to the development of advanced data-driven control methods for complex nonlinear autonomous systems.