Current attempts to explore and persistently monitor the oceans beneath the ice sheets are limited by the absence of highly-reliable Autonomous Underwater Vehicles (AUVs) for long-duration operations. Underwater vehicles routinely fail in these environments due to collisions, drifting ice passages and actuator faults. The challenge consists in designing reliable underwater platforms that pair mechanical resiliency with a lightweight embedded software model that can react to unforecast situations.

SUBIceSAT, an AI-powered software model will drive a constellation of under-ice AUVs: neural control functions will be synthesised to preserve stability of the vehicle dynamics in case of unforeseen situations, without requiring power hungry continuous monitoring sensing. A robust fault-tolerant guidance and control architecture will be designed, capable of rescheduling online mission objectives, exploiting reduced actuator sets. The learning of the guidance and control model will be driven by neural Lyapunov Functions and their correctness will be certified by formal verification methods.

Formally-correct AUV navigation unlocks long-duration under-ice ocean sensing of ice-shelf melt, supporting the Forecasting Tipping Points' early-warning thesis by developing new sensing systems and driving rapid deployment for new technologies. SUBIceSat will allow pairing above-ice data collection systems with persistent under-ice in-situ monitoring, ultimately generating new datasets to enrich melting modelling and prediction.

Davide holds a PhD in control systems for autonomous underwater vehicles from UCL, and a BSc and MSc in control engineering from Politecnico di Milano. He has been a Postdoctoral Fellow at UCL since 2023 where he has been managing the UCL Ocean Towing Tank. Prior, he worked in industrial R&D at INESC-TEC and Airbus Defence and Space. His research focuses on hardware design of autonomous maritime vehicles, on fault-tolerant control systems and on formally-correct machine learning-based control methods.