Deep Reinforcement Learning Applied for Threat Engagement by Loyal Wingmen Drones

• Davi Guanabara De Aragão ITA
• Marcos R. O. A. Maximo ITA
• José Fernando Basso Brancalion EMBRAER

Abstract

This study explores using Loyal Wingmen drones in counter- UAV systems to detect and neutralize multiple aerial threats cooperatively. We employ Curriculum Learning and Bayesian Optimization to enhance the Proximal Policy Optimization algorithm, which trains a Deep Neural Network to perform high-level decision-making. The effectiveness of this network is assessed in simulations that pit it against an increasing number of Loitering Munitions, with a comparative analysis of a Behavior Tree agent. A key finding is the effective deployment of simulated LiDAR technology for dynamic multi-target detection, which surpasses traditional systems limited to observing a fixed number of entities.

Keywords: Laser radar, Weapons, Conferences, Education, Decision making, Deep reinforcement learning, Vehicle dynamics, Robots, Optimization, Drones, Machine Learning, Neural Network, Unmanned Aerial Vehicle, Air Defense System