The main interest of our research group is in model predictive control, as one of the tools in optimization problem solving, and in machine learning techniques, as a support mechanism in different decision making processes. Ultimately, our group aims at developing an optimization framework in which these two approaches are combined, in order to get more efficient and more effective solutions in more complex and uncertain scenarios.

Some of our past work includes model predictive control in unit commitment problem solving on an array of multiple hydropower plants, as well as control of specific processes in thermo and solar power plants. Other examples include medical applications, like feedback control of the glucose level in humans.  Another very specific area of interest is mobile vehicle motion planning. As part of my PhD research I used model predictive control to address the problem of mobile ground vehicle motion planning on rough terrains. The research resulted in a novel MPC-based optimization framework capable of navigating the vehicle through both known and unknown rough terrains. In order to use the MPC optimization framework on terrains of a larger scale, an appropriate cost map was developed forming terminal costs within the objective function. The initial work was performed in cooperation with the Politecnico di Milano DEIB Merlin research group and the robotics group at JPL/NASA. Currently, an extensions to the algorithm is being developed and is performed in collaboration with the Merlin research group. Another novel coverage planner for multi-vehicle navigation planning, intended to cover a given region of interest in an efficient and coordinated manner, was developed as part of a research stay at Imperial College London. Our group now adapted this algorithm in constrained coverage problems by using model predictive control.

When it comes to machine learning applications, some of the techniques were used in addressing problems of extracting brain signals with a low signal to noise ratio. As an example, an approach based on Principal component analysis combined with Independent component analysis was developed in cooperation with Airlab at Politecnico di Milano to improve a P300 speller device. Other experience involves industrial applications in working with Bayes classifiers, logistic regression, neural networks and other classification and regression tools, while solving some decision making and control problems. Neural networks in cruise control were implemented and the results validated on an Audi A8 in Ingolstadt in cooperation with the Friedrich-Alexander-Universität Erlangen-Nürnberg.

Currently, our research group is working on ideas in the field of nonlinear control theory, in combining model predictive control and some available approximate nonlinear control techniques such as SDRE control. Other ideas involve the use of model predictive control supported by sample based techniques such as RRT (rapidly exploring random trees), a well known tool in the robot motion planning community. Ultimately, our research group is aiming at developing a novel chance constrained model predictive control framework by using classification techniques. Current research involves as well work on the development of a fault tolerant based planner for a quadcopter, to solve the underwater coverage problem by using an underwater vehicle. The focus thereby is to find an appropriate cooperative control framework to guide the aerial vehicle while it physically interacts with the underwater vehicle. The work is done as part of the MORUS project.