Real-Time Validation of Machine Intelligence Controlling Unmanned Vehicle Autonomous Operations

To realize the full potential of autonomous systems, it is imperative that they behave safely, correctly, ethically, and legally. Providing these assurances through offline verification alone is insufficient, due to the complex and changing nature of autonomous systems. Online monitoring and corrective actions are necessary to account for uncertainties, and to increase trust between a human supervisor and the system. Verus Research therefore proposes VERETAS, the VErification in REal-time for Trust in Autonomous Systems tool. VERETAS achieves the five essential elements for an effective online verification tool: Speed, communication, changing requirements, predictive capabilities, and corrective actions. First, it integrates real-time software capabilities for reactive monitoring into a graphical user interface designed to simply and effectively communicate with the operator in a manner that increases trust in the system. It handles complex and changing system specifications through a requirements analysis tool. Finally, it combines formal verification with machine learning to predict future system behaviors and provides corrective actions to mitigate violations before or immediately after they occur. Anticipating future violations before they occur enables preventative actions and drastically increases the safety and trust in the autonomous system, and hence the utility of VERETAS.