Designing reliable and adaptive control for autonomous systems
Maico Engelaar defended his PhD thesis at the Department of Electrical Engineering on January 28.
Control design is a cornerstone of modern engineering, shaping technologies that range from industrial robots and power grids to drones, vehicles, and everyday navigation systems. As autonomous systems take on increasingly complex responsibilities, they must make decisions, coordinate with others, and handle uncertainty without human intervention. In his PhD research, Maico Engelaar explores how to design controllers that meet these demands while remaining both reliable and scalable.
Autonomous systems often operate in environments where circumstances change quickly. New tasks may appear during operation, and multiple systems may need to work together. At the same time, real鈥憌orld data is noisy, incomplete, or unpredictable, meaning that many control problems are inherently stochastic. Maico Engelaar developed provably correct control methods that address these challenges simultaneously by enabling cooperation between agents, allowing systems to adapt to new tasks on the fly, and handling uncertainty in a mathematically rigorous way.
Enhancing adaptability through stochastic model predictive control
A key part of this research focuses on stochastic model predictive control (SMPC), a framework widely used for real鈥憈ime decision鈥憁aking under uncertainty. First, Engelaar introduces a risk鈥慳ware SMPC method that gives autonomous systems flexibility while guaranteeing that the system operates reliable and that the controller remains feasible at all times. By tuning the degree of flexibility, a careful balance between adaptability and assurance is achieved. He then extends this method to allow systems to take on tasks introduced during execution, using temporal logic to specify what must be achieved and when. Finally, he expands the approach to multi鈥慳gent systems, where tasks are allocated intelligently and agents coordinate their actions while still respecting shared rules. Finally, the approach is extended to systems with multiple, potentially heterogeneous agents, in which tasks are allocated and coordinated using clever allocation mechanisms. This enables multiple systems to cooperate simultaneously, respond flexibly to changes, and still operate correctly in accordance with agreed-upon rules.
Keeping complex systems manageable
To ensure the scalability of the proposed methods, Engelaar developed two additional techniques to reduce system complexity. The first technique considers uncertainties described as mixtures of Gaussian distributions (a commonly used probability distribution for modeling uncertainty) and reformulates them into simpler probability distributions for which solutions are available in the literature. The second technique reduces the amount of information the system must process simultaneously, thereby simplifying the handling of uncertainties. Both techniques ensure that controllers designed for a simplified system can be safely and correctly applied to the original, complex system.
Simulations and examples
The effectiveness of these techniques is demonstrated through academic simulations and examples, both for single-agent systems, such as vehicles or power controllers, and for systems with multiple cooperating agents, such as a network of shuttle vehicles. These results show that the proposed methods can introduce both adaptability and scalability while maintaining reliability and correctness.
Title of PhD thesis: . Supervisors: Dr. Sofie Haesaert and dr. Mircea Lazar.