In the last decades, the advancement in computing hardware as well as the significant improvements in developing numerical solvers for convex optimization problems have gradually increased the interest of the control community towards the applications of MPC to different challenging scenarios. The disruptive success of MPC controllers in several applications is mainly ascribable to their peculiarity of providing optimized performance while guaranteeing mission and system constraints satisfaction via repetitive online optimization. This unique ability represents an appealing quality for those scenarios and applications in which autonomy, safety, reliability, cost-saving and flexibility are crucial requirements. MPC uses an explicitly dynamic plant model to predict the effect of future reaction of the manipulated variables on the output. Then, an optimal input trajectory is computed based on the predictions by minimizing a pre-defined finite horizon cost function. At each time step, such a finite horizon optimization problem is solved, and only the first input is applied to the system in a receding horizon fashion. When MPC is applied to linearized dynamics with linear constraints, the optimization problem can be reduced to a constrained quadratic programming (QP) problem. On the other hand, the use of nonlinear models in some other MPC applications is motivated by the possibility of improving control by improving the quality of the prediction. However, in most real applications, exact models can not be obtained, either due to model uncertainty or external disturbances. In particular, the uncertainty is assumed to be bounded, the problem is most commonly addressed in literature by formulating a robust MPC (RMPC) problem, where the control law is computed to satisfy the constraints for every possible uncertainty realization. This pessimistic view may imply inherent conservativeness and significantly degrade the overall controller performance. A valid alternative is offered by stochastic MPC (SMPC) approaches, where the constraints are interpreted probabilistically, allowing for a small violation probability. In this framework, even unbounded uncertainty can be studied. The main drawback is the computational burden, which could limit the real-time application of these schemes. All the above mentioned approaches share common features and drawbacks, e.g. computational burden, which can be properly combined with the needs claimed by real complex scenarios, transferring the benefits of these control techniques to specific applications. This has been made possible by the significant amount of theoretical and algorithmic results that have been proposed in the literature in the last decade. Moreover, technological improvements allowed to have access to devices with more powerful computational capability, able to solve fairly complex optimization problems in real-time.
Aim of this workshop is to present state-of-the-art applications of model predictive control (MPC), because of its long tradition of success as a very powerful and versatile advanced control technique and the growing interest shown by industry in a large variety of application domains. Emphasis will be given to highlighting the benefits of exploiting this advanced control techniques and the main challenges tackled when innovative and promising theoretical approaches have to be combined and have to comply with application requests and limitations. The range of applications that will be presented in the workshop includes virtual vehicles and driving simulators, personalized medical treatment, attack detection in cyber-physical systems, unmanned spacecraft maneuvers, Unmanned Aerial Vehicles (UAVs) for agriculture 4.0, autonomous driving. During the workshop, the attendee will be driven through several thrilling and promising applications, each one exploiting a different MPC techniques, ad-hoc tailored for complying with applied domain needs and available computation capabilities. The list of potential speakers and a brief abstract of the presented topics are provided in the following section.
The target audience includes graduate students and researchers in control, robotics, computer scientists, physicists and engineers working on the modeling and control of uncertain systems. The topics covered in the talks of this session will be particularly useful to students and researchers working on stochastic optimal control, stochastic model predictive control, and trajectory optimization problems for uncertain systems (e.g., autonomous robotic systems operating in dynamic environments).
Institute of Electronics, Computer and Telecommunication Engineering
Italian National Research Council
Rio Tinto Centre for Mine Automation, Australian Centre for Field
Robotics, The University of Sydney, Sydney