簡易檢索 / 詳目顯示

回結果列表
研究生： 洪浚洋
HUNG, JUYN YANG
論文名稱： 資源解耦異質SLAM融合之強健室內導航與動態地圖維護系統
Resource-Decoupled Heterogeneous SLAM Fusion for Robust Indoor Navigation with Dynamic Map Maintenance
指導教授： 江凱偉
Chiang, Kai-Wei
學位類別： 碩士
Master
系所名稱： 工學院 - 測量及空間資訊學系
Department of Geomatics
論文出版年： 2026
畢業學年度： 114
語文別： 英文
論文頁數： 138
中文關鍵詞： 異質多機器人系統 、跨平台地圖傳輸 、位姿圖檢查點序列化 、Google Cartographer SLAM 、自適應蒙地卡羅定位 、時間彈性帶局部規劃器 、資源解耦 、動態地圖維護 、室內導航
外文關鍵詞： Heterogeneous Multi-robot Systems, Cross-Platform Map Transfer, Pose-Graph Checkpoint Serialization, Google Cartographer SLAM, Adaptive Monte Carlo Localization, Timed Elastic Band, Resource Decoupling, Dynamic Map Maintenance, Indoor Navigation
相關次數： 點閱：2  下載：0 
分享至:
查詢本校圖書館目錄 查詢臺灣博碩士論文知識加值系統 勘誤回報

    • 在無 GPS 的室內環境中，當先驗地圖須跨異質機器人平台共享時，會面臨一項根本挑戰：運動所引致之姿態擾動、感測器安裝高度與感測量程之差異，決定了由某一平台所測繪之地圖是否仍為另一平台可用之有效定位參考。本論文提出並實驗驗證一套資源解耦之異質 SLAM 融合框架，將跨平台地圖移轉之成立條件表述為可判定的準則，而非針對單一組載具以經驗方式逐次試出。
    本框架依能力向量將各平台指派為測繪代理人或導航代理人，該向量涵蓋地形可及性、感測量程、續航與運動姿態擾動。跨平台移轉係以可重載之位姿圖檢查點為單位，而非柵格化之占據網格，藉此保留原會話之約束結構，使異質平台上所取得之軌跡得以透過跨軌跡回環約束進行聯合再優化。本文所稱「融合」，係指不同平台之軌跡併入同一位姿圖並接受聯合優化，而非各平台間即時交換狀態。兩平台之相容性於部署前即由三項條件判定：掃描平面高度差須落於環境之垂直同質區間內、測繪端與導航端之量程比不得低於一、以及檢查點之序列化須於時基仍推進時完成。
    本框架以 Puppy Pi 四足機器人為測繪代理人、TurtleBot3 輪式平台為導航代理人，於真實室內環境中完成實例化。實測掃描平面高度差為 3.8 公分、有效量程比為 2.29，均滿足上述條件。輪式平台所錄製之測繪會話隨後併入源自四足機器人之位姿圖，產生同時包含兩平台軌跡之單一優化圖；經目視檢查，合併後之地圖牆面幾何與僅由四足所建者一致，未出現註冊失敗特有之牆面結構重疊。於所得先驗地圖上之導航採用自適應蒙地卡羅定位與時間彈性帶局部規劃器，且導航階段無須四足機器人在場。雙層空間表示架構支援執行期地圖維護，區分以射線投射清除之暫態障礙物，與觸發移轉流程重新執行之永久結構變化。
    閉環「返回原點」實驗橫跨三種幾何複雜度遞增之場景，於 50 公尺軌跡上記錄之最大平移均方根誤差為 4.92 公分，相對誤差約為 0.10%。僅由導航平台單獨測繪所得之地圖，其匯出量體遠小於四足所得者，反映角色分派所依據之量程不對稱性。實驗結果確認：由受步態姿態擾動影響之平台所測繪之先驗地圖，足以支撐異質平台達成公分級導航精度；且此類移轉之成立條件，得以表述為可適用於本研究所採載具以外之形式。

    Indoor navigation in GPS-denied environments presents a fundamental challenge when a prior map must be shared across heterogeneous robotic platforms, since differences in locomotion-induced attitude disturbance, sensor mounting height and sensing range determine whether a map surveyed by one platform remains a valid localization reference for another. This thesis proposes and experimentally validates a resource-decoupled framework for heterogeneous SLAM fusion, in which the conditions governing cross-platform map transfer are stated as decidable criteria rather than discovered empirically for a single platform pairing.
    The framework assigns each platform to the role of surveying agent or navigating agent according to a capability vector comprising terrain accessibility, sensing range, endurance and locomotion-induced attitude disturbance. Cross-platform transfer is performed on a reloadable pose-graph checkpoint rather than a rasterized occupancy grid, so that the constraint structure of the originating session is preserved and trajectories acquired on dissimilar platforms can be jointly re-optimized through cross-trajectory loop-closure constraints. The term fusion is used here to denote the merging of trajectories into a common pose graph subject to joint optimization, and not the real-time exchange of state between agents. Compatibility between two platforms is determined prior to deployment by three conditions: the scan-plane height offset must lie within a vertically homogeneous interval of the environment, the ratio of surveying to navigating sensing range must not fall below unity, and serialization of the checkpoint must complete while the time base is still advancing.
    The framework was instantiated on a Puppy Pi quadruped acting as the surveying agent and a TurtleBot3 wheeled platform acting as the navigating agent, operating in a real indoor environment. The measured scan-plane offset was 3.8 cm and the effective range ratio was 2.29, both satisfying the stated conditions. A mapping session recorded on the wheeled platform was subsequently merged into the pose graph originating from the quadruped, yielding a single optimized graph containing trajectories from both platforms; visual inspection of the merged map showed wall geometry consistent with that of the quadruped-only map, with no duplication of wall structure of the kind a failed registration produces. Navigation on the resulting prior map employed Adaptive Monte Carlo Localization and the Timed Elastic Band local planner, and required no presence of the quadruped during the navigation phase. A dual-layer spatial representation supports runtime map maintenance, distinguishing transient obstacles, which are cleared by ray casting, from persistent structural change, which triggers re-execution of the transfer procedure.
    Closed-loop return-to-origin trials across three scenarios of increasing geometric complexity recorded a maximum translational RMSE of 4.92 cm over a 50 m trajectory, corresponding to a relative error of approximately 0.10%. A map surveyed by the navigating platform alone was substantially smaller in exported extent than the one obtained by the quadruped, reflecting the range asymmetry from which the role allocation follows. The results confirm that a prior map surveyed by a platform subject to gait-induced attitude disturbance supports centimeter-level navigation on a dissimilar platform, and that the conditions under which such transfer succeeds can be stated in a form applicable beyond the particular platforms employed here.

    中文摘要 I ABSTRACT II ACKNOWLEDGEMENTS IV TABLE OF CONTENTS V LIST OF TABLES XI LIST OF FIGURES XII CHAPTER 1. INTRODUCTION 1 1.1 Background and Literature Review 1 1.2 Motivation, Objectives and Contributions 5 1.2.1 Research Motivation 5 1.2.2 Research Objectives 6 1.2.3 Research Contributions 7 1.3 Thesis Organization 10 CHAPTER 2. FUNDAMENTALS AND RELATED TECHNOLOGIES 13 2.1 Sensor Principles and Specifications 13 2.1.1 2D LiDAR Operating Principles 13 2.1.2 Inertial Measurement Unit (IMU) 14 2.2 Software Environment 15 2.2.1 Ubuntu 20.04 LTS Operating System 15 2.2.2 ROS Noetic Middleware Architecture 16 2.3 ROS Communication and Coordinate Systems 18 2.3.1 ROS Communication and Modular Topology 18 2.3.2 Transform Tree (TF) and Geometric Coordinate Frames 19 2.4 Simultaneous Localization and Mapping (SLAM) 21 2.4.1 Theoretical Foundation of SLAM 21 2.4.2 Google Cartographer SLAM 22 2.5 Occupancy Grid Map Representation 24 2.5.1 Probabilistic Occupancy Grid Theory 24 2.6 Adaptive Monte Carlo Localization (AMCL) 26 2.6.1 Probabilistic Particle Filter Pipeline 26 2.6.2 KLD-Sampling and Computational Resource Efficiency 27 2.7 Timed Elastic Band (TEB) Local Planner 30 2.7.1 Costmap 2D Architecture and Inflation Layers 30 2.7.2 TEB Trajectory Optimization Theory 31 CHAPTER 3. A GENERALIZED FRAMEWORK FOR HETEROGENEOUS ROBOT INTEGRATION 34 3.1 From Platform-Specific Integration to a Transferable Framework 34 3.1.1 Limitations of Platform-Specific Heterogeneous Integration 34 3.1.2 Design Objectives and Scope 35 3.2 Problem Formalization 36 3.2.1 Agents and Capability Vectors 36 3.2.2 Sessions, Checkpoints, Overlap and Handoff 36 3.2.3 Problem Statement 37 3.3 The Resource-Decoupled Heterogeneous Mapping-Navigation Handoff Framework 38 3.3.1 Architectural Overview 38 3.3.2 L1: Role Allocation 40 3.3.3 L2: Sensor Compatibility 41 3.3.4 L3: Frame and Namespace Contract 43 3.3.5 L4: Map Handoff 44 3.3.6 L5: Runtime Map Maintenance 45 3.3.7 L6: Validation Protocol 46 3.4 Deployment Procedure and Feasibility Checking 47 3.4.1 Deployment Workflow 47 3.4.2 Compatibility Checking Procedure 49 3.4.3 Failure Modes and Fallback Strategies 51 3.5 Generality of the Framework 52 3.5.1 Instantiation across Platform Pairings 52 3.5.2 Applicability Boundaries 54 3.5.3 Relation to Concurrent and Decentralized Multi-Robot SLAM 55 3.6 Summary 56 CHAPTER 4. INSTANTIATION OF THE FRAMEWORK ON A QUADRUPED–WHEELED PAIRING 57 4.1 Platform Characterization and Role Allocation 57 4.1.1 Available Platforms 57 4.1.2 Capability Vectors and Role Assignment 58 4.2 Sensing Configuration and Compatibility Verification 59 4.2.1 Sensor Specifications and Effective Range 59 4.2.2 Scan-Plane Geometry 60 4.2.3 Blind Bands 61 4.2.4 Choice of Ranging over Optical Sensing 62 4.3 SLAM Back-End Selection 63 4.3.1 Evaluation Criteria 63 4.3.2 Candidate Algorithms 64 4.3.3 Selection Rationale 67 4.4 Estimator Configuration 68 4.4.1 Configuration as an Instantiation of the Interface Parameters 68 4.4.2 Absorption of Gait-Induced Attitude Disturbance 70 4.4.3 Pose Graph Optimization Parameters 71 4.5 Frame and Namespace Realization 73 4.5.1 Coordinate Frame Chains 73 4.5.2 Realization of the Frame Contract 74 4.6 Handoff Execution 75 4.6.1 Session Structure 75 4.6.2 Checkpoint Serialization 77 4.6.3 Cross-Platform Merge 78 4.6.4 Outcome of the Merge 79 4.7 Runtime Maintenance Realization 81 4.8 Checklist Results 82 4.9 Summary 83 CHAPTER 5. VALIDATION OF THE INSTANTIATED FRAMEWORK 84 5.1 Experimental Design 84 5.1.1 Hardware Configuration 84 5.1.2 Testbed Layout and Dimensions 88 5.1.3 Navigation Routes and Path Lengths 89 5.1.4 Accuracy Assessment Method 91 5.2 Robust Trajectory Planning in Complex Static Environments 92 5.2.1 Scenario A: Navigation through a Narrow Structural Bottleneck 93 5.2.2 Scenario B: Dynamic Trajectory Adjustment in Continuous Sharp Turns 95 5.3 Dynamic Map Maintenance and Topological Updates 99 5.3.1 Feature Injection: Adding New Obstacles 99 5.3.2 Map Healing: Clearing Outdated Obstacles 101 5.3.3 Topological Updates: Adapting to Permanent Structural Changes 102 5.4 High-Precision Navigation Closed-Loop and Relative Error Assessment 103 5.4.1 Quantitative Analysis of Closed-Loop Navigation Performance 103 5.5 Summary 107 CHAPTER 6. CONCLUSION AND FUTURE WORK 108 6.1 Conclusion 108 6.2 Limitations 109 6.2.1 Terrain advantage was not experimentally validated 109 6.2.2 Mapping accuracy was not measured against an external reference 110 6.2.3 Angular measurement uncertainty is comparable to the smallest reported error 110 6.2.4 Two-dimensional representation and the residual blind band 110 6.3 Future Work 111 6.3.1 Quantitative validation of the handoff gain 111 6.3.2 Autonomous promotion of persistent structural change 111 6.3.3 Extension to three-dimensional representation 111 6.3.4 Instantiation on a handheld surveying agent 112 6.3.5 Deployment without continuous connectivity 112 REFERENCES 113 APPENDIX 117

    Agarwal, S., Mierle, K., & The Ceres Solver Team. (2022). Ceres solver (Version 2.1). Retrieved from http://ceres-solver.org/
    Amsters, R., & Slaets, P. (2019). TurtleBot 3 as a robotics education platform. In Robotics in Education (pp. 170-181). Springer, Cham. https://doi.org/10.1007/978-3-030-26945-6_16
    Bitelli, G., Forte, A., & Mandanici, E. (2026). SLAM mobile mapping for complex archaeological environments: Integrated above–below-ground surveying. Geomatics, 6(2), 31. https://doi.org/10.3390/geomatics6020031
    Bresenham, J. E. (1965). Algorithm for computer control of a digital plotter. IBM Systems Journal, 4(1), 25–30. https://doi.org/10.1147/sj.41.0025
    Canonical Ltd. (2020). Ubuntu 20.04 LTS (Focal Fossa) release notes. Retrieved from https://releases.ubuntu.com/20.04/
    Cano, J., Molinos, E., Nagarajan, V., & Vijayakumar, S. (2015). Dynamic process migration in heterogeneous ROS-based environments. In 2015 International Conference on Advanced Robotics (ICAR) (pp. 518–523). IEEE. https://doi.org/10.1109/ICAR.2015.7251505
    D'Angelo, S., Pagano, F., Caccavale, R., Scognamiglio, V., De Crescenzo, A., Merone, P., Ciaravino, S., Finzi, A., & Lippiello, V. (2026). H-CoRE: A cooperative framework for heterogeneous multi-robot exploration and inspection. Drones, 10(4), 232. https://doi.org/10.3390/drones10040232
    Durrant-Whyte, H., & Bailey, T. (2006). Simultaneous localization and mapping (SLAM): Part I. IEEE Robotics & Automation Magazine, 13(2), 99–110. https://doi.org/10.1109/MRA.2006.1638022
    Elfes, A. (1989). Using occupancy grids for mobile robot perception and navigation. Computer, 22(6), 46–57. https://doi.org/10.1109/2.30720
    Farahsari, P. S., Farahzadi, A., Rezazadeh, J., & Bagheri, A. (2022). A survey on indoor positioning systems for IoT-based applications. IEEE Internet of Things Journal, 9(10), 7680–7699. https://doi.org/10.1109/JIOT.2022.3149048
    Foote, T. (2013). tf: The transform library. 2013 IEEE Conference on Technologies for Practical Robot Applications (TePRA), 1–6. https://doi.org/10.1109/TePRA.2013.6556373
    Fox, D. (2003). Adapting the sample size in particle filters through KLD-sampling. The International Journal of Robotics Research, 22(12), 985–1003. https://doi.org/10.1177/0278364903022012001
    Fox, D., Burgard, W., Dellaert, F., & Thrun, S. (1999). Monte Carlo Localization: Efficient position estimation for mobile robots. Proceedings of the Sixteenth National Conference on Artificial Intelligence (AAAI-99), 343–349. Retrieved from https://aaai.org/papers/050-aaai99-050-monte-carlo-localization-efficient-position-estimation-for-mobile-robots/
    Grisetti, G., Kümmerle, R., Stachniss, C., & Burgard, W. (2010). A tutorial on graph-based SLAM. IEEE Intelligent Transportation Systems Magazine, 2(4), 31–43. https://doi.org/10.1109/MITS.2010.939925
    Grisetti, G., Stachniss, C., & Burgard, W. (2007). Improved techniques for grid mapping with Rao-Blackwellized particle filters. IEEE Transactions on Robotics, 23(1), 34–46. https://doi.org/10.1109/TRO.2006.889486
    Hess, W., Kohler, D., Rapp, H., & Andor, D. (2016). Real-time loop closure in 2D LiDAR SLAM. 2016 IEEE International Conference on Robotics and Automation (ICRA), 1271–1278. https://doi.org/10.1109/ICRA.2016.7487258
    Hiwonder. (2023). PuppyPi quadruped robot with AI vision user manual. Retrieved from https://www.hiwonder.com/products/puppypi
    Huang, C., & Liu, R. (2022). Design attention awareness among robots for uncertainty-adaptive heterogeneous teaming. 2022 IEEE International Conference on Advanced Robotics and Its Social Impacts (ARSO), 1–6. https://doi.org/10.1109/ARSO54254.2022.9802958
    International Organization for Standardization. (2024). Robotics — Performance criteria and related test methods for service robots — Part 2: Navigation (ISO 18646-2:2024). https://www.iso.org/standard/82643.html
    Jo, D., & Kwon, Y. (2025). Generation of critical information and sharing mechanism for multi-robot mission success. IEEE Access, 13, 146855–146873. https://doi.org/10.1109/ACCESS.2025.3600319
    Kakhkharov, A., Kim, J.-W., & Choi, J.-H. (2026). Quadruped robots in construction automation: A comprehensive review of applications, localization, and site-level operations. Buildings, 16(5), 962. https://doi.org/10.3390/buildings16050962
    Kohlbrecher, S., von Stryk, O., Meyer, J., & Klingauf, U. (2011). A flexible and scalable SLAM system with full 3D motion estimation. 2011 IEEE International Symposium on Safety, Security, and Rescue Robotics, 155–160. https://doi.org/10.1109/SSRR.2011.6106777
    LDROBOT. (2022). LD19 LiDAR sensor datasheet. Retrieved from https://www.ldrobot.com/
    Li, T., Sun, S., & Sattar, T. P. (2013). Adapting sample size in particle filters through KLD-resampling. Electronics Letters, 49(12), 740–742. https://doi.org/10.1049/el.2013.0296
    Lu, D. V., Hershberger, D., & Smart, W. D. (2014). Layered costmaps for context-sensitive navigation. 2014 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 709–715. https://doi.org/10.1109/IROS.2014.6942636
    Macenski, S., & Jambrecic, I. (2021). SLAM toolbox: SLAM for the dynamic world. Journal of Open Source Software, 6(61), 2783. https://doi.org/10.21105/joss.02783
    Moravec, H. P. (1989). Sensor fusion in certainty grids for mobile robots. In A. Casals (Ed.), Sensor devices and systems for robotics (pp. 253–276). Springer. https://doi.org/10.1007/978-3-642-74567-6_19
    Open Robotics. (2021). ROS Noetic Ninjemys documentation. Retrieved from http://wiki.ros.org/noetic
    Quigley, M., Conley, K., Gerkey, B., Faust, J., Foote, T., Leibs, J., Wheeler, R., & Ng, A. Y. (2009). ROS: an open-source Robot Operating System. ICRA Workshop on Open Source Software, 3(3.2), 5. https://ai.stanford.edu/~ang/papers/icraoss09-ROS.pdf
    ROBOTIS. (2023). TurtleBot3 e-manual. Retrieved from https://emanual.robotis.com/docs/en/platform/turtlebot3/overview/
    Rösmann, C., Feiten, W., Wösch, T., Hoffmann, F., & Bertram, T. (2012). Trajectory modification considering dynamic constraints of autonomous robots.
    Proceedings of the 7th German Conference on Robotics (ROBOTIK 2012), 1–6. VDE. https://api.semanticscholar.org/CorpusID:2043153
    Rösmann, C., Hoffmann, F., & Bertram, T. (2017). Kinodynamic trajectory optimization and control for car-like robots. 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 5681–5686. https://doi.org/10.1109/IROS.2017.8206458
    Seki, H., Yamamoto, Y., & Nagasawa, S. (2024). The influence of micro-hexapod walking-induced pose changes on LiDAR-SLAM mapping performance. Sensors, 24(2), 639. https://doi.org/10.3390/s24020639
    Thrun, S., Burgard, W., & Fox, D. (2005). Probabilistic robotics. MIT Press.
    Zafari, F., Gkelias, A., & Leung, K. K. (2019). A survey of indoor localization systems and technologies. IEEE Communications Surveys & Tutorials, 21(3), 2568–2599. https://doi.org/10.1109/COMST.2019.2911558
    Zhao, Y., Smith, J. S., Karumanchi, S. H., & Vela, P. A. (2020). Closed-loop benchmarking of stereo visual-inertial SLAM systems: Understanding the impact of drift and latency on tracking accuracy. In 2020 IEEE International Conference on Robotics and Automation (ICRA) (pp. 1105–1112). IEEE. https://doi.org/10.1109/ICRA40945.2020.9197003

    QR CODE