Session 2.1: Methodological Challenges

Session 2.1.A: Learning and Data-driven control​

Abstract Probably the biggest development in control in the past decade, and one of the most important moving forward, is the incorporation of machine learning enabled components into the feedback control loop. These developments are becoming game-changers in a variety of application areas, such as robotics, transportation, health care, manufacturing, infrastructure, agriculture. A central aspect in data driven control is the relationship between computations done off-line and on-line. Other important aspects are: How should data be collected for training? To what extent can experimental data be replaced by simulations? Is it possible to formally certify learning based controllers for use in safety-critical applications?

Session 2.1.B: Safety-critical systems​

Description: TBA

Session 2.1.C: Resilience, security, privacy ​

Description: TBA

Session 2.1.E: Architecture and control​

Abstract The societal systems discussed thus far are composed of interconnections of interacting elements such as sensors, actuators, computers, communication devices, human machine interfaces, humans, algorithms, and software. 'Control architecture' describes how the system components are connected and how they interact. There is a growing awareness that architectural decisions play an important role in putting control technologies into real-world societal systems.  While virtually all contemporary engineering systems of moderate complexity (e.g., aviation, processor design, autonomous-vehicles) take a top-down hierarchical and layered approach to architecture design, today’s societal-scale systems (e.g., internet, power-grid, social networks, policy) are the result of an evolutionary bottom-up and ad-hoc design, resulting in unforeseen, cryptic, and at times catastrophic failures. The goal of this section is to initiate the study of an integrated, unifying framework of architecture and of control as a `control architecture.’  While this remains a long-term objective, we point to methods that show some progress in this challenging research area.