Schedule

"Lesson Learned from Using Overarching Properties in Industrial Case Studies"

Zamira Daw, Senior Manager, AI Systems Engineering Team Lead, Raytheon Technologies Research Center

Aerospace systems are seeing a strong trend towards ‘electronification’ and increased reliance on software. This move towards increasingly critical and complex functionality and safety behaviors captured in aerospace electronics is driving consideration of new certification methods. Systems and software complexity is challenging certifying agencies, approving organizations, academia and industry to search for new and innovative technologies to provide effective assurance of safety-based systems. Some of the challenges in the certification of AI components are non-determinism, difficulty tracing to detailed requirements, black box algorithmic complexity, and pre-existing AI/ML packages (OS, COTS, non-aerospace applications). Overarching Properties (OPs) have been created by an international working group and are being evaluated by the National Aeronautics and Space Administration (NASA), the Federal Aviation Administration (FAA), industry, and other certifying agencies in an effort to streamline certification processes. The hope is that the FAA may eventually establish an Advisory Circular that offers the OPs as a Means of Compliance (MoC). In this talk, we discuss the challenges of current certification processes and show some industrial case studies that uses OPs as means of compliance.

“Computer-Aided Generation of Assurance Cases”

Pierluigi Nuzzo, Assistant Professor, Department of Electrical and Computer Engineering and Computer Science, University of Southern California

Timothy Wang, Principal Research Engineer, Raytheon Technologies Research Center

Safety-critical products are required to satisfy domain-specific certification standards to ensure a safe, secure, and reliable system. In the aerospace and medical device industries, we have seen a growing interest in the usage of assurance cases in place of certification standards, providing a structured argument on why a product is safe or sufficiently secure for a given application. Therefore, there is a need for methodologies and tools that support the generation of correct and compelling assurance cases. Major challenges stem from the complexity of realistic designs, the heterogeneous nature of the required evidence, the need to assess the quality of an argument, and the need to produce interpretable outcomes. In this paper, we present an assurance case generation framework that automatically constructs assurance cases based on a system specification, assurance evidence, and domain expert knowledge, captured in assurance patterns. A commercial aerospace case study shows that the generated assurance cases are meaningful, and performance experiments show the efficiency and scalability of the approach.

"Run-time monitoring for safe robot autonomy"

Marco Pavone, Associate Professor, Department of Aeronautics and Astronautics, Stanford University

In this talk I will present our recent results towards designing run-time monitors that can equip any pre-trained deep neural network with a task-relevant epistemic uncertainty estimate. I will show how run-time monitors can be used to identify, in real-time, anomalous inputs and, more broadly, provide safety assurances for learning-based autonomy stacks. Finally, I will discuss how run-time monitors can also be used to devise effective strategies for data lifecycle management.

"Ariadne: A common-sense thread for enabling provable safety in air mobility systems with unreliable components"

Eric Feron, Professor, Division of Computer, Electrical and Mathematical Sciences and Engineering, King Abdullah University of Science and Technology (KAUST)

"Certification and Proof of Collision Avoidance Using Active Corners"

Jean-Baptiste Jeannin, Assistant Professor, Department of Aerospace Engineering at the University of Michigan — Ann Arbor

"Provably Correct Neurosymbolic Controllers for Assured Autonomous Systems"

Yasser Shoukry, Assistant Professor, Department of Electrical Engineering and Computer Science, University of California — Irvine