Collective Meta-Reasoning Runtime Assurance of Machine Intelligence for Unmanned Autonomous Vehicles

Barron Associates has teamed with a prominent researcher in the field of formal verification of cyber physical systems to propose a new paradigm in runtime assurance for complex autonomous systems controlled and operated by artificial machine intelligence. A two-stage approach is considered in which formal verification processes are first performed offline at algorithm design time. Online, during operation, centralized and decentralized systems perform meta-reasoning of real-time decision processes to validate decision outcomes at a level greater than what each system can reason alone. This adds robustness to unanticipated missing information, communication dropouts, knowledge omissions, and potential malicious cyber-attacks. In Phase I, a challenge problem will be constructed to aid in the development of the proposed approach. A specific mission function will be identified as a technology demonstrator and the benefits of the approach will be proven in simulation.