Deep Reinforcement Learning-Based Cooperative Survivability Maximization for a UAV Fleet on an Air-to-Ground Mission

This study focuses on the cooperative strategy development of a UAV team that operates in a hostile environment in which the radar and weapon systems try to track and eliminate them. To simulate the hostile defense system, we present Markov models that generate the detecting and tracking probabiliti...

Full description

Saved in:
Bibliographic Details
Published inHavacılık ve Uzay Teknolojileri Dergisi Vol. 15; no. 2; pp. 94 - 107
Main Author Barış Başpınar
Format Journal Article
LanguageEnglish
Published Turkish Air Force Academy 01.07.2022
Online AccessGet full text

Cover

More Information
Summary:This study focuses on the cooperative strategy development of a UAV team that operates in a hostile environment in which the radar and weapon systems try to track and eliminate them. To simulate the hostile defense system, we present Markov models that generate the detecting and tracking probabilities of a radar system, and calculate the multiple-shot survivability of air vehicles that fly within the hostile environment. A cooperative strategy development procedure is presented based on proximal policy optimization algorithm, which is a deep reinforcement learning method. It is shown that the UAV team can develop cooperative strategies by exploiting enemy’s weakness to maximize team survivability in an air-to-ground mission after training with the proposed reinforcement learning scheme.
ISSN:1304-0448