Modeling of Intelligent UAV Reconnaissance Behaviors and Decisions Based on Virtual Simulation

This paper proposes a method for UAV firefield reconnaissance behavior and decision-making modeling based on imitation learning. Expert wears a VR headset and use a flight joystick to operate a UAV in a virtual environment to complete the task of firefield reconnaissance. The experts' operations are recorded and collected as sample data for training intelligent UAV through imitation learning. For the problem of random and complex firefield reconnaissance, multiple imitation learning algorithms are adopted to establish an intelligent UAV firefield reconnaissance model. Through the evaluation of the model, the algorithm based on behavioral cloning and generative adversarial algorithm is selected as the learning algorithm. By adding external rewards and updating training parameters, a generalized intelligent UAV firefield reconnaissance behavior and decision-making model is obtained, which can complete the random firefield reconnaissance task safely and timely.