Intention recognition of UAV swarm with data-driven methods

Abstract

UAVs have been increasingly used in military and commercial applications. The theory of UAV swarm behavir has gradually matured and moved to the real application stage. Fast and accurate recognition of the intentions of UAV swarms become a key part of dealing with coming swarms. This paper proposes a data-driven approach to realize the recognition of the typical intentions of UAV swarm. The UAV swarm’s intention is divided into three basic categories: expansion, free movement, and contraction. The dubins model is introduced to depict and study the dynamic characteristics of the movement of the UAV swarm. Simulation experiments are performed through software to collect data and to verify and refine the proposed data-driven intention recognition approach. Moreover, real flight experiments are conducted to test the feasibility and accuracy of the proposed approach, from which key steps about the neural network building and training for intention recognition have been summarized, and satisfying results in intention recognition with high accuracy and stability during the entire movement of the UAV swarm have been achieved.