Cyber-Physical Systems: Theoretical and Practical Challenges

ubiquitous embedded systems employed to monitor and control physical processes: cars, airplanes, automotive highway systems, air traffic management , etc. In the past, research on embedded systems tended to focus on the design optimization problems of these computational devices. In recent years, the focus has shifted towards the complex synergy between the computational elements and the physical environment with which they interact. The term Cyber-Physical Systems (CPS) was coined to refer to such interactions. In CPS, embedded computation and communication devices, together with sensors and actuators of the physical sub-stratum, are federated in heterogeneous, open, systems-of-systems. Examples include smart cities, smart grids, medical devices, production lines, automotive controllers, and robotics. Here we discuss some of the research problems that we are addressing, on both the theory and practical application scenarios of CPS, such as automotive and medical systems. Theoretical foundations and challenges The behaviour of CPS is characterized by the nonlinear interaction between discrete (computing device) and continuous phenomena (the physical sub-stratum). For this reason, research on hybrid systems plays a key role in modelling and analysing CPS. CPS are usually spatially distributed and they exhibit emergent behaviours (i.e. traffic jams, cyber-attacks), which result from interactions among system components, and which cease to exist when specific components are removed from the systems. Owing to their ubiquity and impact on every aspect of our life, one of the greatest challenges of this century is to efficiently predict the emergent behaviours of these systems. The complexity of their models, however, often hinders any attempt to exhaustively verify their safe behaviour. An alternative method is to equip CPS with monitors and to predict emergent behaviours at runtime. This approach makes CPS self-aware, opening up new approaches to designing systems that can dynamically reconfigure themselves in order to adapt [1] to different circumstances. However, monitoring introduces a runtime overhead that may alter the timing-related behaviour of the system under scrutiny. In applications with real-time constraints, overhead control strategies may be necessary to reduce the overhead to acceptable levels by, for example, turning on and off the monitoring. Gaps in monitoring, however , introduce uncertainty in the monitoring results. Hence, our current research [1] also focuses on efficient techniques to quantify this uncertainty and compute an estimate of the current state of the system. Automotive scenario The extensive integration of sensor networks and computational power into automotive systems over recent years has enabled the development …