Explainable and Fast-Converging Artificial Intelligence Solution to Control a Nonlinear Aircraft Model in Air Combat

This paper presents an algorithm for chasing target aircraft in air combat scenarios, focusing on explainability and safety. Unlike conventional approaches utilizing reinforcement learning, our method employs a problem-specific neural network architecture with one hidden layer, trained online to track the desired path and heading angle in a 3D environment. The algorithm distinguishes between offensive and defensive modes, selecting optimal positions for the tracker aircraft and controlling it accordingly. We introduce a different training procedure where the neural network learns from the system responses without labeled output information, ensuring quick convergence and explainability. Through simulations, we demonstrate the reliability and effectiveness of our algorithm and neuro-controller structure with the help of decision tree structure in air-to-air combat tasks.