New framework improves real-time robot monitoring for maritime safety

Robotic systems must operate safely under uncertainty, but existing monitoring methods demand extensive data or explicit uncertainty models. Elizabeth Dietrich and colleagues from UC Berkeley and NTNU propose a new framework that leverages data-driven reachable sets—geometric bounds on possible system states—to evaluate complex specifications in real time. This approach avoids the need for full uncertainty distributions, making it far more data-efficient and practical for deployment.

The framework is instantiated for maritime navigation, where vessels must follow complex traffic rules (COLREGs). The authors built a data-efficient pipeline to construct reachable sets and derived a monitoring formulation suitable for real-time execution. Simulation and hardware experiments showed robust performance under realistic disturbances, achieving improved risk detection compared to state-of-the-art metrics. This marks a step toward safer autonomous ships and could extend to other robotic domains needing reliable specification monitoring.