This multi-disciplinary research is at the intersection of control engineering and computer science. Specifically, methods from control and dynamical systems theory are being used to address challenges in real-time embedded systems. The central objective of the research is to develop next generation verification algorithms for real-time embedded systems. The modeling and analysis framework for determining robustness of embedded systems is based on the stochastic robustness framework, where the binary notion of robustness is discarded and the notion of a risk-adjusted robustness margin tradeoff is adopted. Key elements in this verification-based research include modeling of parametric uncertainty in embedded systems and development of computational tools for accurate prediction of uncertainty in real-time systems; robust performance analysis of various scheduling algorithms; and impact of various models of computations (anytime, imprecise, interlaced) on system level robustness. Uncertainty propagation tools will use methods based on Monte-Carlo techniques, polynomial chaos, and transfer operators (such as Perron-Frobenius and Koopman operators). Tools developed in the area of multi-rate robust control will be used to analyze the effect of various models of computation on robustness of the system. It is expected that successful completion of this project will enable accurate assessment of reliability of embedded systems across a wide array of engineering disciplines. This research project will also train a new generation of researchers with a mixed background in dynamical systems, control theory, and computer science. This is aligned with current and future research and industrial needs.
Due to the increasing role of information based systems, it is expected that future realtime embedded control systems will be large, complex, distributed and dynamic. Pervasive computing, sensing and communication will be common in such systems. It will be absolutely necessary for control and embedded system engineers to conduct formal analysis to establish reliability of such dynamic realtime systems by guaranteeing timely execution of software elements in the presence of uncertainty. We identify characterization of uncertainty and the analysis of its evolution, in such large scale distributed interacting embedded systems, as a key research challenge in proving reliability of future embedded systems. In this project we study the effect of computational uncertainty in the performance of control systems that ensure safe operation of above described engineering systems. We propose a paradigm shift in the verification of embedded systems. The focus is on quantifying the uncertainty in distributed embedded systems due to uncertain transient computational overloads. The interaction is modeled as a Markov jump system and performance is characterized by a suitable metric in the space of probability density functions. A new randomized scheduler is proposed that optimally tradeoffs between computational time and robust system performance. We expect our efforts to have a broad impact at several levels. Our proposed framework will enable accurate assessment of reliability of embedded systems across a wide array of engineering disciplines. Currently, by various accounts, reliability assessment of an embedded system is the most serious bottleneck that engineers face in delivering reliable embedded systems. The research will result in analysis tools that will be provided to the embedded systems community, including academia and industry. One of the general goals of this multidisciplinary project is to help create a new generation of researchers with a mixed background in dynamical systems, control theory and computer science. This is aligned with current and future needs of the industry. It is expected such a training is going to expand employment opportunities for the students. Moreover, we plan to integrate the results of our research to develop relevant course work at Texas A&M University, for wider dissemination of knowledge. Our work from this effort has been published in reputed journals and presented in conferences. A list of material from this work is available in the project webpage. Students who worked in this project include: Abhishek Halder and Kooktae Lee.