Cyber-physical systems (CPSs) allow computer systems to monitor and control the physical world in a new way that could revolutionize many areas of science and engineering. However, they are often too complex for non-specialists to use. The aim of this work is to develop new technology to manage this complexity, enabling scientists and engineers to use CPSs just like other tools and instruments. This research takes a comprehensive approach to macroprogramming -- the task of programming an entire network of devices as a single, programmable substrate. This research exploits global, network-wide information about a CPS provided by a macroprogram to improve traditional software engineering techniques such as testing, debugging, analysis, and optimization. New techniques are being developed that use global information to optimize system performance, automatically generate test cases, and reduce the state space for analysis. This work is developing new programming abstractions that allow the separation of the application logic from quality-of-service requirements and hardware requirements, improving code portability and reuse. This research will produce a comprehensive development environment for CPSs called MacroLab. The new tools developed will greatly simplify the process of their programming and make them more accessible to non-experts. By taking a holistic view of the network and its software, MacroLab will manage a range of complex, interacting issues that would be extremely difficult to manage by hand. MacroLab will be tested pilot studies, including environmental monitoring. A graduate CPS course will be developed. MacroLab will be used for course experiments and in senior capstone projects.

This CPS Frontiers project addresses highly dynamic Cyber-Physical Systems (CPSs), understood as systems where a computing delay of a few milliseconds or an incorrectly computed response to a disturbance can lead to catastrophic consequences. Such is the case of cars losing traction when cornering at high speed, unmanned air vehicles performing critical maneuvers such as landing, or disaster and rescue response bipedal robots rushing through the rubble to collect information or save human lives. The preceding examples currently share a common element: the design of their control software is made possible by extensive experience, laborious testing and fine tuning of parameters, and yet, the resulting closed-loop system has no formal guarantees of meeting specifications. The vision of the project is to provide a methodology that allows for complex and dynamic CPSs to meet real-world requirements in an efficient and robust way through the formal synthesis of control software. The research is developing a formal framework for correct-by-construction control software synthesis for highly dynamic CPSs with broad applications to automotive safety systems, prostheses, exoskeletons, aerospace systems, manufacturing, and legged robotics. The design methodology developed here will improve the competitiveness of segments of industry that require a tight integration between hardware and highly advanced control software such as: automotive (dynamic stability and control), aerospace (UAVs), medical (prosthetics, orthotics, and exoskeleton design) and robotics (legged locomotion). To enhance the impact of these efforts, the PIs are developing interdisciplinary teaching materials to be made freely available and disseminating their work to a broad audience.

The project investigates a formal verification framework for artificial pancreas (AP) controllers that automate the delivery of insulin to patients with type-1 diabetes (T1D). AP controllers are safety critical: excessive insulin delivery can lead to serious, potentially fatal, consequences. The verification framework under development allows designers of AP controllers to check that their control algorithms will operate safely and reliably against large disturbances that include patient meals, physical activities, and sensor anomalies including noise, delays, and sensor attenuation. The intellectual merits of the project lie in the development of state-of-the-art formal verification tools, that reason over mathematical models of the closed-loop including external disturbances and insulin-glucose response. These tools perform an exhaustive exploration of the closed loop system behaviors, generating potentially adverse situations for the control algorithm under verification. In addition, automatic techniques are being investigated to help AP designers improve the control algorithm by tuning controller parameters to eliminate harmful behaviors and optimize performance. The broader significance and importance of the project are to minimize the manual testing effort for AP controllers, integrate formal tools in the certification process, and ultimately ensure the availability of safe and reliable devices to patients with type-1 diabetes. The framework is made available to researchers who are developing AP controllers to help them verify and iteratively improve their designs. The team is integrating the research into the educational mission by designing hands-on courses to train undergraduate students in the science of Cyber-Physical Systems (CPS) using the design of AP controllers as a motivating example. Furthermore, educational material that explains the basic ideas, current challenges and promises of the AP concept is being made available to a wide audience that includes patients with T1D, their families, interested students, and researchers. The research is being carried out collaboratively by teams of experts in formal verification for Cyber-Physical Systems, control system experts with experience designing AP controllers, mathematical modeling experts, and clinical experts who have clinically evaluated AP controllers. To enable the construction of the verification framework from the current state-of-the-art verification tools, the project is addressing major research challenges, including (a) building plausible mathematical models of disturbances from available clinical datasets characterizing human meals, activity patterns, and continuous glucose sensor anomalies. The resulting models are integrated in a formal verification framework; (b) simplifying existing models of insulin glucose response using smaller but more complex delay differential models; (c) automating the process of abstracting the controller implementation for the purposes of verification; (d) producing verification results that can be interpreted by control engineers and clinical researchers without necessarily understanding formal verification techniques; and (e) partially automating the process of design improvements to potentially eliminate severe faults and improve performance. The framework is evaluated on a set of promising AP controller designs that are currently under various stages of clinical evaluation.