Forum 360: Safety Certification for Machine Learning 6 August 2021 1130 - 1245
Machine learning is going to be necessary for autonomy. Currently, we rely on a human to adapt to new situations, but that won’t be the case for an uncrewed vehicle. As machine learning is impossible to prove 100% correct, it is difficult to certify. Two workarounds have been suggested: wrapping machine learning in safe watchdog systems that will allow it to be guarded in a way to make it safe and safe learning work. Hear experts discuss their research into these approaches and migrating them through the processes necessary for approval and certification.
Associate Professor, Aeronautics and Astronautics, Stanford University
Fellow, Collins Aerospace
Eric N. Johnson
Professor of Aerospace Engineering, Pennsylvania State University (PSU), and Director, PSU UAS Research Laboratory (PURL)
Senior Vice President and General Manager, Boeing Program, Spirit AeroSystems, Inc.
Subproject Manager, NASA System-Wide Safety Project, and Assurance of Responsible Automation Technical Lead, Advanced Air Mobility Project, NASA Langley Research Center
Associate Technical Fellow, Flight Controls & Autonomy, Sikorsky, a Lockheed Martin Company