Designed and developed a multi-agent reinforcement learning system to extend the world’s first AI fighter pilot to command a fleet of heterogeneous collaborative aircraft;
Trained prototype agents for 2 vs 2 dogfights, thoroughly testing the simulation, training, and visualization pipelines across various objectives in performance, flight safety, and human trust;
Helped Shield AI survive a significant down-select over Lockheed Martin and Boeing, culminating in real-life flights on a modified F-16 with no human interventions.