| 摘要: | 本論文主旨為設計與實現一個適用於室內環境的自主移動機器人(Autonomous Mobile Robot, AMR)，其能夠自主行走、自動導航及避障，主要應用於校園室內環境的配送服務，如送餐和送件。同時，針對無人車的導航準確性進行深入性的研究，探索並比較多感測器融合技術在不同演算法下的性能表現，進而優化其導航與定位的精準度。
本論文首先利用光學雷達(LiDAR)結合同步定位與地圖構建(Simultaneous Localization and Mapping, SLAM)方法對整個實驗場域進行建圖，隨後對於系統中選用的感測器做資料前處理，消除雜訊獲取更精確的姿態訊息。此外，為確保不同感測器資料在時間上的一致性，使用資料時間同步技術，減少了延遲與融合誤差，也能準確地反映無人車即時的姿態，為多感測器融合奠定了良好的基礎。在完成上述準備工作後，分別運用擴展卡爾曼濾波(Extended Kalman Filter, EKF)和無跡卡爾曼濾波(Unscented Kalman Filter, UKF)兩種演算法針對單一感測器和多感測器進行無人車狀態預測，執行自主導航與避障任務。本論文的重點在於透過實驗數據，分析不同演算法及感測器配置下的導航精準度表現，並調適這兩個種演算法中的參數設置，以提升其預測精準度，最終選擇平均誤差最小的方案作為實際應用。另外，本研究還探討了其他優化策略，進一步提升導航系統的精準度。
本研究在建立於Linux操作系統，使用機器人作業系統(Robot Operating System, ROS)作軟體開發框架實驗，利用TCP/IP協議進行數據傳輸與通訊，並在硬體部分以NVIDIA Jetson AGX Xavier做為系統核心，無人車底盤則採用ZW3840差速輪架構，實現軟硬結合的協作設計，完成複雜的指派任務。本論文的實驗結果顯示，在20公尺的導航路徑上，系統的平均導航誤差僅為0.13公尺，相當於1%左右的相對誤差，此一研究成果可顯示自主機器人在校園室內環境中的自主導航與避障的可行性，並有潛力應用於各種配送服務。
;The main objective of this thesis is to design and implement an Autonomous Mobile Robot (AMR) for indoor environments, capable of autonomous walking, navigation, and obstacle avoidance. The system is primarily applied to campus delivery services, such as food and document delivery. This study also investigates the navigation accuracy of AMR by exploring and comparing multi-sensor fusion performance under different algorithms to optimize navigation and localization precision.
LiDAR combined with Simultaneous Localization and Mapping (SLAM) maps the experimental environment, followed by sensor data preprocessing to eliminate noise and improve pose accuracy. Data synchronization techniques ensure consistency across sensor data, reducing delay and fusion errors while accurately reflecting the robot′s real-time state. Two algorithms, Extended Kalman Filter (EKF) and Unscented Kalman Filter (UKF), are applied to single and multi-sensor configurations for vehicle state prediction, enabling autonomous navigation and obstacle avoidance. Experimental data is analyzed to evaluate navigation accuracy under different configurations, with parameters adjusted to enhance prediction precision. The algorithm with the smallest average error is selected for implementation, while additional optimization strategies are explored to improve system accuracy further.
The system is built on the Linux operating system using Robot Operating System (ROS) for software development and TCP/IP for data communication. NVIDIA Jetson AGX Xavier is the system′s core, and the ZW3840 differential drive chassis provides the hardware foundation. Experimental results show the system achieves an average navigation error of 0.13 meters over a 20-meter path, corresponding to a relative error of about 1%. These findings demonstrate the feasibility of autonomous navigation and obstacle avoidance in indoor campus environments and highlight its potential for different delivery services. |
