Publications

Detailed Information

Persistent surveillance using multiple robots: coordination and path planning : 다중 로봇을 이용한 지속적 정찰 및 감시 기법 연구: 조정 및 경로계획

DC Field Value Language
dc.contributor.advisor김현진-
dc.contributor.author김우진-
dc.date.accessioned2017-07-13T06:16:08Z-
dc.date.available2017-07-13T06:16:08Z-
dc.date.issued2014-08-
dc.identifier.other000000021157-
dc.identifier.urihttps://hdl.handle.net/10371/118404-
dc.description학위논문 (박사)-- 서울대학교 대학원 : 기계항공공학부, 2014. 8. 김현진.-
dc.description.abstract다중 로봇 시스템은 각각의 로봇이 개별적으로, 혹은 한 대의 로봇으로 해결하기 힘든 작업을 수행할 수 있다. 본 학위논문은 복수 로봇의 지속적인 감시와 관련하여 (1) 지속적인 환경 정보 수집, (2) 환경 정보의 모델링, 그리고 (3) 환경 현상의 경계선 추종을 위한 다중 로봇 시스템의 조정 기법을 개발한다. 능동적인 환경 정보 수집 문제는 정보를 효과적으로 얻기 위해 환경 정보의 불확실성을 감소시키는 로봇의 경로를 생성하는 것이다. 본 연구에서는 간헐적 칼만 필터가 적용하여 정보의 질을 보장할 수 있는 로봇의 목표 방문 분포를 구한다. 또한, 목표 방문 분포를 만족하는 경로를 생성하기 위하여 distribution-motivated A* 알고리듬을 제안한다. 제안된 알고리듬은 목표 방문 분포를 달성하는 경로를 생성한다. 수집된 환경정보를 모델링하기 위하여, support vector machine (SVM) 기법의 앙상블 조합법을 제안한다. 각 로봇이 각자의 데이터를 이용하여 지역 모델링을 수행하고 이들의 앙상블 조합을 통해서 기존의 전역 SVM에 비하여 정확도와 성능 측면에서 더 좋은 결과를 얻을 수 있으며, 데이터량이 크거나 통신의 제약에 대해서 유연한 솔루션을 제공한다. 또한, 환경 현상의 경계선 추종을위하여, 복수 이동형 로봇의 제어기를 설계한다. 환경정보 모델링을 통하여 얻어진 초고차원 결정 함수를 이용하여 목표로하는 경계를 향하면서 동시에 그 주위를 순환하는 벡터 필드를 생성하고, 복수 로봇 집단의 배열이 등간격을 유지하며 경계를 둘러싸도록 하는 조정기법을 개발한다. 앞에서 제안된 기법들을 검증하기 위하여, 비행로봇과 지상로봇의 수치 시뮬레이션과 실험을 수행한다. 시뮬레이션 결과를 통하여 능동형 환경 정보 수집 문제의 정보 수집 효
율성과, 앙상블 SVM의 정확도 그리고 복수 로봇의 경계 추종 기법의 성능을 검증한다. 더욱이 실험적 결과를 통하여 실제 이종 로봇 시스템의 지속적인 감시에 제안된 기법의 적용 가능성을 확인한다.
-
dc.description.abstractA multi-agent system is a distributed system composed of multiple interacting intelligent agents within an environment. Multi-agent systems can be used to solve problems that are difficult or impossible for an individual agent or a single system to solve. Especially, the heterogeneity of the multi-agent system may be due to physical difference between agents or behavioral difference when robots serve diverse roles in a cooperating team. In this dissertation, the coordination and path planning methods of the heterogeneous robots are proposed for persistent surveillance in the following three parts: (i) persistent robotic environmental sensing, (ii) modeling environmental information, and (iii) coordination of multi-robot systems for boundary tracking.
A key problem of robotic environmental sensing and monitoring is that of active sensing. This problem can be solved by generating the most informative observation paths for the robots, to limit the uncertainty in modeling and predicting an environmental phenomenon. In this work, the intermittent Kalman filter is applied to the active sensing. From the Kalman process, the uncertainties of the sensed information for each cell of the discrete workspace are obtained. And the target distribution of the robots visit which guarantees the predefined performance limit is calculated. After that, the robot is guided to visit each cell of the workspace according to the target distribution by a modified A* algorithm which we call distribution-motivated A*. The proposed algorithm generate paths achieving the target distribution.
Using the sensed information, the ensemble implementation of support vector machine (SVM) is proposed to model the environmental phenomenon in a distributed manner. The ensemble combination is a modeling or prediction technique that builds sub-predictors for each robot with its own dataset, and combines the sub-predictors with proper weights for the final prediction. This work shows that a well-organized collection of sub-predictors yields a more accurate result when compared with conventional SVM predictors. Moreover, this technique offers a flexible solution for the problems arising from large data and communication limitation.
This work also considers a boundary tracking problem using mobile agents, in which a controller is designed for the mobile agents to move along the boundary of the environmental phenomenon. By using the hyper-dimensional decision function obtained from the SVM described in the previous paragraph, the hyper-potential field is constructed to generate a velocity vector field which is globally attractive to the desired closed path with circulation at the desired speed. The collective configuration of the multiple robots is coordinated in the evenly-spaced formation that encloses the boundary by minimizing the level of synchrony of the agents.
In order to validate the proposed methods, the simulations and the experiments with multiple flying and ground robots are carried out. The simulation results verify guaranteed informativeness during the active sensing, accuracy and robustness of the ensemble SVM, and good performance of the collective boundary tracking. Furthermore, the experimental results demonstrate the applicability of the proposed methods to the practical heterogeneous robotic systems for the persistent surveillance.
-
dc.description.tableofcontents1 Introduction
1.1 Motivation
1.2 Literature Review
1.2.1 Persistent Active Sensing
1.2.2 Modeling the Environmental Information
1.2.3 Boundary Tracking with Multi-Robots
1.3 Objectives and Contributions
1.3.1 Active Sensing
1.3.2 Modeling the Environmental Information
1.3.3 Boundary Tracking with Multi-Robots
1.4 Thesis Orgarnization
2 Persistent Active Sensing of Environmental Phenomena
2.1 Research Objectives
2.2 Problem Description
2.2.1 Uncertainty Representation
2.3 Proposed Active Sensing Algorithm
2.3.1 Stability of Kalman Filter
2.3.2 Target Distribution and Required Agent Number
2.3.3 Persistent and Informative Path Generation
2.4 Extension toMulti-agent Systems
2.5 Simulation Results
3 Modeling Environmental Information
3.1 Problem Description
3.2 Support Vector Machine (SVM)
3.2.1 Linear Support Vector Machines
3.2.2 Nonlinear Support Vector Machines
3.2.3 One-Class Support Vector Machine
3.3 Subpredictors and Ensemble Combination
3.3.1 Training Subpredictors: Local Ensemble
3.3.2 Aggregation Ensemble
3.4 Weighted Ensemble
3.4.1 Known Variances
3.4.2 Unknown Variances
3.5 Performance Analysis
3.5.1 If the Variance of the Noise is Known
3.5.2 If the Variance of the Noise is Unknown
3.6 Algorithm Validation
4 Environmental Boundary Tracking
4.1 Problem Description
4.2 SVM-based Lyapunov Vector Field
4.2.1 SVM-based Potential
4.2.2 Lyapunov-based Vector Field
4.2.3 Vector Filed Construction
4.2.4 Analysis on ∇ϕ=0
4.3 Coordination of Multiple Mobile Agents
4.4 Simulation Results
5 Experimental Validations and Results
5.1 Experimental Setup
5.2 Experimental Results
5.2.1 Active Sensing with Flying Robots
5.2.2 Boundary Tracking using Ground Robots
5.2.3 Integrated System
6 Conclusions
-
dc.formatapplication/pdf-
dc.format.extent3335553 bytes-
dc.format.mediumapplication/pdf-
dc.language.isoen-
dc.publisher서울대학교 대학원-
dc.subject지속적 감시기법-
dc.subject능동형 환경 정보 수집-
dc.subject환경 정보 모델링-
dc.subject경계선 추종-
dc.subject.ddc621-
dc.titlePersistent surveillance using multiple robots: coordination and path planning-
dc.title.alternative다중 로봇을 이용한 지속적 정찰 및 감시 기법 연구: 조정 및 경로계획-
dc.typeThesis-
dc.description.degreeDoctor-
dc.citation.pages111-
dc.contributor.affiliation공과대학 기계항공공학부-
dc.date.awarded2014-08-
Appears in Collections:
Files in This Item:

Altmetrics

Item View & Download Count

  • mendeley

Items in S-Space are protected by copyright, with all rights reserved, unless otherwise indicated.

Share