- 學位論文
適用於田間機器人之改良式同步定位與地圖建構演算法
An Improved Simultaneously Localization and Mapping Algorithm for Field Robots
摘要
本論文提出一適用於農業環境且無需外部基礎設施之自主式同步定位與地圖建立 (Simultaneously Localization And Mapping, SLAM) 機器人系統。本系統以自行開發之改良式同步定位與地圖建立演算法，使用二維雷射測距儀獲取三維空間資訊，故能克服農業環境單調且多變的特殊性，並增加應用彈性及減少系統建置成本。此外，本系統引入具調控者機制之行為與感測器控制系統，以使機器人於未知環境中進行自主導航與探索行為。為實現上述系統，本研究針對農業環境開發一大小適中，且可克服崎嶇地貌的機器人做為系統平台。為進行空間掃描，二維雷射測距儀係以傾斜向地面的方式架設於機器人上。配合機器人本體的移動，即可獲得連續之環境距離資料。所得之資料將轉換為空間中之點集，再依其高度資訊進行空間分層，同層資料視為一組資訊。利用粒子濾波器 (particle filter) 的方法將各組掃描資料分層做結合，即可同步求得環境場景地圖與機器人位置之估計結果。移動掃描過程中，地圖資料將於一定時間間隔分離為新的地圖區塊，以避免環境中雜亂資料的不必要累積，進而加強定位與地圖建立之可信度。經農業環境實驗，驗證此系統能不受農業環境中充斥之雜亂與細碎物體干擾，而有效處理農業環境機器人定位與地圖建立的問題。機器人能於未知地圖環境中實現自動導航功能，到達目的地之平均誤差小於35公分，平均每步費時6.1秒。若是預先載入環境地圖，則平均誤差小於40公分，平均每步費時2.5秒。
並列摘要
In this thesis, we propose a robot system with SLAM (Simultaneously Local- ization and Mapping) applied to agricultural environment which does not need the auxiliary equipment. To adapt the characteristically monotone and diverse agricul- tural environment, this system uses a two-dimensional laser range nder to get the three-dimensional information based on the improved SLAM algorithm we devel- oped. Such method can enhance the application exibility, as well as to reduce the system construction cost. Moreover, this system includes the superior mechanism to supervise the sensor and behavior control part for robot to autonomously navigate and explore an unknown environment. In order to realize the system mentioned above, we implement a robot appro- priate for the agricultural environment as the system platform. This robot, which is suitable in size, can overcome rugged topography. To progress spatial scanning, the two-dimensional laser range nder set up on the robot has a tilt to the ground. While the robot moving, the continuous distance information of the environment will be obtained at the same time. These obtained data will be transformed into point cloud and divided into multi-layers. The points of each layer will be combined to a "layer map" by particle lter. By the map from every layer, the robot's pose and environment state will be known. When the robot scanning the space, the map will be separated into several sub-maps to avoid the clutter accumulating and to en- hance the localization and mapping results. Con rmed by experimental result, the system can ignore the disturbance from the agricultural environment and solve the robot simultaneously localization and mapping problem. The SLAM experimental results demonstrate that the average errors of eld robot position is less than 35 cm; very step takes 6:1 seconds. If the map has been constructed, the average error is less than 40 cm; every step takes 2:5 seconds. The result is feasible for agricultural applications.