CPS: Small: Software-State Observability in CPS
Lead PI:
Jason Rife
Co-PI:
Abstract
Cyber-physical system (CPS) technologies, such as automated aircraft and cars, have become sufficiently complex that CPS software verification is now a major bottleneck in product development. This project examines new approaches for auto-generating reduced models of CPS software, in order to incorporate those models in analysis, for instance, in system-wide simulations or bug detection. This project will allow CPS software to be adapted and analyzed much more flexibility in comparison with state-of-the-art methods, which limit software developers by prohibiting use of many modern programming constructs and by penalizing iterative software improvements during the design process. The project's intellectual merit is the introduction of a theory of software-state observability, which will have wide utility for CPS analysis including in applications such as online bug detection. To this end the project concentrates on three specific aims: (i) the development of concepts for reduced-order software modeling based on static and dynamic analyses of CPS software programs, (ii) the formulation of a theory of software-state observability to enable state estimation across the boundaries of physical and software components, and (iii) the application of these theories to online bug monitoring for an open-source flight control system. The project represents a fundamental departure from the conventional treatment of software in a CPS, where software must be tightly specified in advance, where the program must be carefully verified to prove that it meets specifications, and where after final validation the software is assumed to be essentially free of bugs. Our approach permits developers much greater latitude in creating new CPS software by requiring reasonable but not excessive initial testing, as justified by better analysis tools and by online reliability monitoring. The result will enable more sophisticated and lower cost automated cars and unmanned aircraft. Results related to the project will be shared through archival publications and data will be made available online through Tufts University at https://tufts.box.com/s/aq305785bg2s3j29lls8u4wdpybzpcth. Software uploaded to this repository will be available under suitable licensing models (such as BSD or Apache). Data will be retained at least through the duration of this project.
Performance Period: 01/01/2019 - 12/31/2021
Institution: Tufts University
Sponsor: National Science Foundation
Award Number: 1836942