Grants and Contributions:

Grant or Award spanning more than one fiscal year. (2017-2018 to 2022-2023)

Objectives: The proposed research is part of a multi-year program to develop numerical algorithms and implementations for the verification, synthesis and design of safe controllers and planners for cyber-physical systems subject to uncertainty arising from the influence of external agents whose behaviour is revealed at run-time. Examples of this often non-random uncertainty include human-in-the-loop control and systems using legacy or independently designed black-box controllers whose safety properties cannot be guaranteed. The techniques developed will maintain safety and support pursuit of the agent's goal in a manner that aligns with user expectations. The research will be informed by shared control applications in powered wheelchairs for cognitively impaired older adults, automated delivery of anesthesia, unmanned aerial vehicles, and collaborative manufacturing.

Approach: The foundation of this research is development of numerical algorithms for approximating reachability of systems with continuous or hybrid state models, and in particular algorithms which can generate results robust to model uncertainty. The goal of this algorithm development is not just verification, but rather characterization of the set of control inputs which lead to correct behaviors. The second element of this research program will be to design interventions that use this information to maintain safety or assist the agent to attain its goal, despite uncertainty about that goal and system state. The affective outcome of these interventions on the system's user will be a key feature in their design

Expected Significance: Unexpected and incorrect operation can lead to a host of negative consequences, and the potential significance of bad behaviours and/or outcomes is increasing as we transition to human-in-the-loop scenarios and/or compose controllers with poorly understood safety characteristics into complex systems. By developing formal verification methods which can treat online input sources in a black-box fashion, my research program will increase the confidence of designers and users that these systems will perform correctly. By considering how these problems are best solved in a diverse set of application areas, approaches which are more broadly applicable can be identified and subsequently extended to other domains in which cyber-physical systems are prominent. Trainees will gain experience in numerical algorithms, robust software development practices, health care technology development and/or robotics, which are all areas with positive job prospects.