Verification and validation of AI systems that control deep-space spacecraft

Abstract

NASA is developing technology for the next generation of deep-space robotic spacecraft, with the aim of enabling new types of missions and radically reducing costs. One technology under development is Autonomy: highly capable spacecraft that perform significant scientific missions with little or no commanding and monitoring from Earth. Artificial Intelligence provides a basis for autonomy technology, but raises issues of verification and validation outside the scope of empirical testing technology for conventionally commanded spacecraft. This paper describes research towards extending formal methods verification techniques for the mathematical verification of AI systems controlling deep-space spacecraft. This paper first overviews a planned space mission called DS-1 which includes an AI-based autonomy experiment. It then describes part of this AI system called the executive, which includes an ‘intelligent’ operating system based on goal-oriented constructs. The paper then describes focused research on applying and extending model-checking technology for verifying both the core services of the executive and the concurrent task programs run by the executive.