Learned Obstacle Warning Assistant using Neurosymbolic AI for Digital Flight Awareness and Survivability against Threats (LOWANDFAST)

Modern anti-aircraft targeting systems are very effective. To counter them flying very close to the ground and obstacles (i.e., hills, trees, wires, and buildings) can block the radar from seeing the aircraft at all, blur its radar signature in ground clutter, and even produce multiple reflections of it—thereby confusing targeting. In short, flying low and fast is critical for Army aircraft survivability. However, even with the best sensors and training, pilots can only react to terrain and obstacles they are cognitively aware of. Pilots are currently limited to what they can see (including with sensors) or know from a map and hazards overlays, but such tools do not have detailed knowledge of the conflict area. This means that in unfamiliar locations, even the best pilots cannot replicate the performance they can achieve when flying in well-understood areas. What is needed is a cognitive terrain flight assistant (CTFA), to provide the pilot with improved awareness of the terrain and obstacles. To build such a system using machine learning algorithms requires addressing three distinct challenges. First, we must capture terrain data at a suitable level of fidelity. We must have access to a digital twin of the combat space at the correct level of fidelity. Second, we must be able to rapidly reason about candidate flight and path options. This planning function must consider the full set of possible options and be sensitive to the capabilities of the specific aircraft and its mission, as well as the terrain and obstacles. This requires a simulation environment that encodes all of these features. Third, we must be able to develop a CTFA that can understand the impacts of terrain and obstacles on possible flight paths and provide both active (“push”) and requested (“pull”) recommendations to the pilot. Such a capability must be accurately and rapidly learned from the captured terrain data in the simulation environment. Only by addressing all three of these challenges can we produce a CTFA that can successfully help human pilots to achieve their most effective low, close, and fast flight in high-threat environments. To address these challenges, Charles River Analytics and University of Iowa’s Operator Performance Laboratory (OPL) propose Learned Obstacle Warning Assistant using Neurosymbolic AI for Digital Flight Awareness and Survivability against Threats (LOWANDFAST). LOWANDFAST combines the OPL’s unparalleled experience and capability in collecting sensor and human performance data in rotorcraft and fixed-wing flight platforms with Charles River Analytics’ neural program policies (NPPs), a neurosymbolic AI approach for deep reinforcement learning that has been integrated into trusted military simulations and learned effective policies at scale. LOWANDFAST will provide pilots with unparalleled awareness of potential hazards and allow increase aircraft survivability in unfamiliar environments.