Skip to main content

Learning, Analysis, SynthesiS and Optimization of Cyber-Physical Systems

Periodic Reporting for period 4 - LASSO (Learning, Analysis, SynthesiS and Optimization of Cyber-Physical Systems)

Reporting period: 2020-05-01 to 2022-04-30

Problem/Issue being addressed & Importance for society
Cyber-physical systems are emerging throughout society in numerous areas including intelligent traffic systems (cars, roads), air traffic control systems (airplanes, airports), smart grid (power plants, wind turbine farms, solar cells), smart buildings (alarm, heating/ventilation/air condition systems, infotainment), and smart cities (buildings, traffic, energy supply, waste water management) as well as medical devices (dialysis machines, pacemakers).

Cyber-physical systems brings the promise of providing better solutions in terms of adaptability, performance, efficiency, functionality and usability. However, cyber-physical systems are often highly safety critical (e.g. medical devices or airplanes), hence utmost care must be taken to ensure that optimization of such performance measures does not violate crucial safety constraints. Given the growth in complexity -- including the growing use of machine learning -- this task becomes increasingly challenging and important for society, with new and scalable methods and tools urgently needed.

Overall objectives
The principle barrier for mastering the engineering of complex CPS being both trustworthy and efficient is the lack of a theoretical well-founded framework for CPS engineering supported by powerful tools, that will enable companies to timely meet increasing market demands. It is the objective of LASSO to provide the new generation of scalable tools for model-based learning, analysis, synthesis and optimization of cyber-physical systems based on a mathematical sound foundation, that enables trade-offs between functional safety and quantitative performance.

LASSO will achieve its objective by ground-breaking and extensive combinations of two different research areas: model checking and machine learning. The framework will develop a complete metric approximation theory for quantitative models, allowing properties to be inferred from reduced or learned models with metric guarantees of their validity in the original system. Further, the applicability of the framework will be validated through a number of CPS case studies, and the tools developed will be made generally available.
The main results achieved so far include the completion of a mathematical and logical foundation for quantitative modelling and reasoning about cyber-physical systems. In particular, a metric theory of “closeness” of models and "degree of satisfaction" has been established. Another line of important results – resulting in two best paper awards – is the development of new efficient datastructures and algorithms significantly improving efficiency of analysis and synthesis for quantitative models. In addition, a combination of symbolic methods and reinforcement learning has been developed and implemented in the tool UPPAAL STRATEGO allowing for synthesis of control strategies that are both safe and optimal. In period UPPAAL STRATEGO has been applied to synthesis of safe and optimal control for a number of case studies, including Home-automation, Adaptive Cruise Control, Intelligent Traffic Lights, Fleet management of drones, and Mission planning for Nano-satellites.
The industrial take-up of our disruptive approach to safe and optimal controller synthesis has progressed much faster than anticipated. In particular, within the domains of Home-Automation and Intelligent Traffic Control progress on making industrial transfer to external partners is being made. Concerning the theoretical foundation -- comprised by a quantitative modelling framework with quantitative (aka metric) verdicts – the work has progressed substantially beyond state-of-the-art and is receiving wide-spread take-up by research colleagues.