HOPPLITE: A Next-Generation Pilot Assistance System

Nap-of-earth (NoE) helicopter flight requires pilots to rapidly interpret their surroundings, anticipate obstacles, and make split-second decisions based on limited information. To maximize survivability, they must fly as close to the terrain as possible—an approach that significantly increases the risk of collision.Advances in the collection of terrain data (from even low-cost and short-lived sUAS) enables centimeter-scale mapping that can capture individual branches and wires. In the near future, one should expect that precise and up-to-date digital models of battlefields, targets, and strategic flight corridors will be available.However, raw data alone does not enhance NoE flight safety. Due to the complexity of vehicle dynamics and environmental effects, the simplified 3D representation traditionally used for path planning does not apply. Therefore, no AI/ML techniques that we are aware of can provide relevant feedback without digitized expert knowledge, which is the exact purpose of our flight model. By marrying high resolution digital twin terrain data with a computerized expert model and modern machine learning algorithms,we can generate clear, actionable routing suggestions—essentially gaining ‘local knowledge’ of the terrain during flight. For example, faced with valley confluence, a cognitive flight assistance model could inform the pilot that the north river valley is safer because it does not have power lines crossing 2 miles ahead, enabling flight at a lower altitude.During a six month Ph. I effort, KEF will develop a machine learning flight guidance model, powered by high-resolution semantic digital twin data. This will lay the groundwork for a Ph. II that leverages state of the art AI/ML techniques to turn this 'offline' flight model into a real-time cognitive terrain pilot assistance system.