DEEP LEARNING METHODS FOR DECENTRALIZED DECISION-MAKING IN COUNTERSWARM ENGAGEMENTS

Simple item page
dc.contributor.advisorKaminer, Isaac I.
dc.contributor.advisorClark, Abram H., IV
dc.contributor.authorSharif, Kurdo A.
dc.contributor.departmentPhysics (PH)
dc.date.accessioned2024-08-19T16:37:43Z
dc.date.available2024-08-19T16:37:43Z
dc.date.issued2024-06
dc.description.abstractThe adoption of unmanned technologies has spurred multidisciplinary research into robotic swarm systems, especially in military contexts. Inspired by biological swarms' problem-solving abilities, these systems offer the advantage of global behavior emerging from local interactions, reducing reliance on centralized control. Traditional approaches to creating emergent behavior in robotic swarms require predictable and controllable swarms with well-defined local rules and complete knowledge of all agents. In counter-swarm engagements, swarm systems need a global strategy that is robust and adaptable to dynamic environments with minimal reliance on complete knowledge. This research investigates an inverse problem: designing local rules to approximate emergent behavior typically based on perfect knowledge and communication by each drone. The objective is to create decentralized regions where a defender drone utilizes a neural network model trained extensively on simulation data. The data, extracted from engagements involving three attackers and one defender, was organized into various input sets representing different features. Post-training regression analysis identified the feature set that generated optimal defender heading angles compared to an oracle algorithm. The results demonstrated that the neural network model optimizes for shorter engagements more effectively than the oracle, validating the feasibility of employing trained networks in lieu of traditional algorithms.en_US
dc.description.distributionstatementDistribution Statement A. Approved for public release: Distribution is unlimited.en_US
dc.description.serviceCaptain, United States Marine Corpsen_US
dc.identifier.curriculumcode533, Applied Physics of Combat Systems
dc.identifier.thesisid39663
dc.identifier.urihttps://hdl.handle.net/10945/73226
dc.publisherMonterey, CA; Naval Postgraduate Schoolen_US
dc.rightsThis publication is a work of the U.S. Government as defined in Title 17, United States Code, Section 101. Copyright protection is not available for this work in the United States.en_US
dc.subject.authorswarmen_US
dc.subject.authorcounter-swarmen_US
dc.subject.authordroneen_US
dc.subject.authorrobotic systemsen_US
dc.subject.authorautonomousen_US
dc.subject.authormachine learningen_US
dc.subject.authorneural networken_US
dc.subject.authordeep learningen_US
dc.titleDEEP LEARNING METHODS FOR DECENTRALIZED DECISION-MAKING IN COUNTERSWARM ENGAGEMENTSen_US
dc.typeThesisen_US
dspace.entity.typePublication
etd.thesisdegree.disciplineApplied Physicsen_US
etd.thesisdegree.grantorNaval Postgraduate Schoolen_US
etd.thesisdegree.levelMastersen_US
etd.thesisdegree.nameMaster of Science in Applied Physicsen_US
relation.isDepartmentOfPublication0966cedc-9b4d-44f6-932d-51786e48444d
relation.isDepartmentOfPublication.latestForDiscovery0966cedc-9b4d-44f6-932d-51786e48444d
Files
Original bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
24Jun_Sharif_Kurdo.pdf
Size:
5 MB
Format:
Adobe Portable Document Format
Download
Collections
Theses