Modeling Autonomous Aerial Vehicle Target Identification Tasks Using Reinforcement Learning

Nate Quirion; Dahai Liu; Andrei Ludu; Mei Ying Lau; Dennis A. Vincenzi

Modeling Autonomous Aerial Vehicle Target Identification Tasks Using Reinforcement Learning

Nate Quirion, Dahai Liu, Andrei Ludu, Mei Ying Lau, Dennis A. Vincenzi

Embry-Riddle Aeronautical University

Research output: Contribution to conference › Presentation

Abstract

Unmanned Aerial Systems (UASs) today are fulfilling more roles than ever before. There is a general push to have these systems feature more advanced autonomous capabilities in the near future. To achieve autonomous behavior it requires some unique approaches for control and decision making. This study used Reinforcement Learning (RL) algorithms to achieve adaptive decision-making capabilities for UASs. An autonomous aerial vehicle was simulated to locate three targets in an unknown area. In order to facilitate the navigation behavior, one navigation agent and one search agent were developed to collaborate in order to meet the requirements of navigating and searching efficiently. Monte Carlo (MC) algorithms and Temporal Difference (TD) algorithms were implemented as the learning methods, their efficiency of navigation and searching were compared. Results were presented and the possible implications to autonomous UAS design were discussed at the end.

Original language	American English
State	Published - Jan 2014
Event	Proceedings of the 2014 Industrial and Systems Engineering Research Conference - Duration: Jan 1 2014 → …

Conference

Conference	Proceedings of the 2014 Industrial and Systems Engineering Research Conference
Period	1/1/14 → …

Keywords

unmanned aerial systems
reinforcement learning
autonomous behavior
decision making
Monte Carlo algorithms
Temporal Difference algorithms

Disciplines

Aerospace Engineering
Computer and Systems Architecture
Digital Communications and Networking

Cite this

@conference{dd8ba698ea194177b10f7c36cfca22b4,

title = "Modeling Autonomous Aerial Vehicle Target Identification Tasks Using Reinforcement Learning",

abstract = " Unmanned Aerial Systems (UASs) today are fulfilling more roles than ever before. There is a general push to have these systems feature more advanced autonomous capabilities in the near future. To achieve autonomous behavior it requires some unique approaches for control and decision making. This study used Reinforcement Learning (RL) algorithms to achieve adaptive decision-making capabilities for UASs. An autonomous aerial vehicle was simulated to locate three targets in an unknown area. In order to facilitate the navigation behavior, one navigation agent and one search agent were developed to collaborate in order to meet the requirements of navigating and searching efficiently. Monte Carlo (MC) algorithms and Temporal Difference (TD) algorithms were implemented as the learning methods, their efficiency of navigation and searching were compared. Results were presented and the possible implications to autonomous UAS design were discussed at the end. ",

keywords = "unmanned aerial systems, reinforcement learning, autonomous behavior, decision making, Monte Carlo algorithms, Temporal Difference algorithms",

author = "Nate Quirion and Dahai Liu and Andrei Ludu and Lau, {Mei Ying} and Vincenzi, {Dennis A.}",

year = "2014",

month = jan,

language = "American English",

note = "Proceedings of the 2014 Industrial and Systems Engineering Research Conference ; Conference date: 01-01-2014",

}

TY - CONF

T1 - Modeling Autonomous Aerial Vehicle Target Identification Tasks Using Reinforcement Learning

AU - Quirion, Nate

AU - Liu, Dahai

AU - Ludu, Andrei

AU - Lau, Mei Ying

AU - Vincenzi, Dennis A.

PY - 2014/1

Y1 - 2014/1

N2 - Unmanned Aerial Systems (UASs) today are fulfilling more roles than ever before. There is a general push to have these systems feature more advanced autonomous capabilities in the near future. To achieve autonomous behavior it requires some unique approaches for control and decision making. This study used Reinforcement Learning (RL) algorithms to achieve adaptive decision-making capabilities for UASs. An autonomous aerial vehicle was simulated to locate three targets in an unknown area. In order to facilitate the navigation behavior, one navigation agent and one search agent were developed to collaborate in order to meet the requirements of navigating and searching efficiently. Monte Carlo (MC) algorithms and Temporal Difference (TD) algorithms were implemented as the learning methods, their efficiency of navigation and searching were compared. Results were presented and the possible implications to autonomous UAS design were discussed at the end.

AB - Unmanned Aerial Systems (UASs) today are fulfilling more roles than ever before. There is a general push to have these systems feature more advanced autonomous capabilities in the near future. To achieve autonomous behavior it requires some unique approaches for control and decision making. This study used Reinforcement Learning (RL) algorithms to achieve adaptive decision-making capabilities for UASs. An autonomous aerial vehicle was simulated to locate three targets in an unknown area. In order to facilitate the navigation behavior, one navigation agent and one search agent were developed to collaborate in order to meet the requirements of navigating and searching efficiently. Monte Carlo (MC) algorithms and Temporal Difference (TD) algorithms were implemented as the learning methods, their efficiency of navigation and searching were compared. Results were presented and the possible implications to autonomous UAS design were discussed at the end.

KW - unmanned aerial systems

KW - reinforcement learning

KW - autonomous behavior

KW - decision making

KW - Monte Carlo algorithms

KW - Temporal Difference algorithms

M3 - Presentation

T2 - Proceedings of the 2014 Industrial and Systems Engineering Research Conference

Y2 - 1 January 2014

ER -