자료유형 | 학위논문 |
---|---|
서명/저자사항 | Artificial Camouflage Systems with Centralized and Decentralized Pattern Formation. |
개인저자 | Li, Yang. |
단체저자명 | University of Colorado at Boulder. Computer Science. |
발행사항 | [S.l.]: University of Colorado at Boulder., 2019. |
발행사항 | Ann Arbor: ProQuest Dissertations & Theses, 2019. |
형태사항 | 111 p. |
기본자료 저록 | Dissertations Abstracts International 81-04B. Dissertation Abstract International |
ISBN | 9781085797566 |
학위논문주기 | Thesis (Ph.D.)--University of Colorado at Boulder, 2019. |
일반주기 |
Source: Dissertations Abstracts International, Volume: 81-04, Section: B.
Advisor: Correll, Nikolaus. |
이용제한사항 | This item must not be sold to any third party vendors. |
요약 | In nature, many animals display the ability to camouflage themselves in the surroundings by adapting their shapes, colors, and textures. How can we engineer artificial systems that have similar ability? Inspired by those natural camouflage, this thesis focuses on exploring and designing the artificial camouflage systems.This thesis first surveys the components needed to design possible systems for camouflage capability, from visual perception, camouflage patterns, pattern formation, to color-changing cells. It then reviews the mathematical models proposed for pattern formation in the literature, most of them based on the reaction-diffusion model. This thesis also surveys the artificial camouflage materials and systems inspired by camouflage animals.A distributed camouflage system for swarm robotics and smart materials is developed and demonstrated with the `Droplet' swarm robotics platform. The system consists of distributed algorithms of dominant color detection, local pattern decision, global pattern consensus, and pattern formation among the swarm. The distributed camouflage system is tested with both simulation and hardware experiments, and the results show that the system can perform camouflage and is robust and scalable.To continue the development of the artificial camouflage systems, two pattern formation algorithms empowered by deep learning are proposed. The first is a decentralized pattern formation with deep reinforcement learning. The second is a centralized pattern formation with generative adversarial networks. Both of them show that they are able to generate the desired patterns for the systems.To provide an efficient camouflage assessment, a 3D simulation is proposed to apply the camouflage patterns on the to-hide objects, and the simulation is followed by object detectors to report the camouflage scores. The images of camouflaged objects are captured with a set of differently configured cameras and passed through a deep-learning based detector to obtain the detection score. Each camera can be set with the distance and angle relative to the object to evaluate if the camouflage is robust to different perspectives. To sum up, this thesis not only provides centralized and decentralized algorithmic foundations for camouflage materials consisting of large numbers of smart particles, but also presents an efficient pipeline for testing artificial camouflage systems. |
일반주제명 | Artificial intelligence. Robotics. Computer science. |
언어 | 영어 |
바로가기 |
: 이 자료의 원문은 한국교육학술정보원에서 제공합니다. |