DC Field | Value | Language |
---|---|---|
dc.contributor.advisor | Shim, Hyunchul | - |
dc.contributor.advisor | 심현철 | - |
dc.contributor.author | Kim, Donghwi | - |
dc.date.accessioned | 2022-04-27T19:31:23Z | - |
dc.date.available | 2022-04-27T19:31:23Z | - |
dc.date.issued | 2021 | - |
dc.identifier.uri | http://library.kaist.ac.kr/search/detail/view.do?bibCtrlNo=963420&flag=dissertation | en_US |
dc.identifier.uri | http://hdl.handle.net/10203/296012 | - |
dc.description | 학위논문(석사) - 한국과학기술원 : 전기및전자공학부, 2021.8,[iv, 41 p. :] | - |
dc.description.abstract | Currently, The main technology of Anti-Drones is to use images to find out the relative position and speed of the target drone and follow it. However, there is no specific maneuver for capturing target drones for multiple agents. This paper proposes element techniques that can subdue the target drone using multiple drones in response to the unauthorized operation and penetration of drones that have recently emerged rapidly. two or three slow-paced pursuit drones were trained to capture one target drone using reinforcement learning in an obstacle environment. multi-agent deep deterministic policy gradients algorithm is used to learn multi-agent. To increase performance using recurrence networks, we apply single-agent off-policy recurrent reinforcement learning methods to multi-agent algorithms to guarantee recurrence and compare and analyze their performance. As a result, storing the hidden state of the recurrent network into the replay buffer and using it in the learning part showed 1.2 times better performance. To maximize performance, we designed the network to share and learn each agent’s experience, increasing the speed by about 2x. | - |
dc.language | eng | - |
dc.publisher | 한국과학기술원 | - |
dc.subject | Deep Reinforcement Learning▼aMulti-Agent Reinforcement Learning▼aRecurrent Network▼aAnti-Drone▼aOff-Policy | - |
dc.subject | 심층 강화학습▼a다중에이전트 강화학습▼a회귀 네트워크▼a안티드론▼a오프 정책 | - |
dc.title | Development of an anti-drone system based on a multi-agent reinforcement learning algorithm | - |
dc.title.alternative | 다중에이전트 강화학습 기반 안티드론 시스템 개발 | - |
dc.type | Thesis(Master) | - |
dc.identifier.CNRN | 325007 | - |
dc.description.department | 한국과학기술원 :전기및전자공학부, | - |
dc.contributor.alternativeauthor | 김동휘 | - |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.