R and D on Complex Autonomous Systems Assurance

Traditional systems have a well-established path to certification. It may be expensive and not as efficient as one might like, but it exists. For example, DO-178C for software. For autonomous systems, and for AI-enabled systems, it is not clear if a path exists. There is a consensus that new assurance techniques are needed, and more importantly, certification guidelines are needed for complex autonomous systems. This presentation reveals an approach to provide recommendations and certification guidelines to a certifying authority for AI-enabled systems. This work was performed through NASA's System-wide Safety Project Technical Challenge 4. The approach includes incorporating careful modifications to industry standards (like ARP4761A and ARP4754B), while engaging certifying authorities (like the FAA) while collaborating with experts in industry, academia and other government agencies to develop tools and techniques that provide assurances to aerospace systems that include complex autonomous systems.