本研究著力於發展可用於田間工作之自主式行走機器人。為達成此目的所開發之系統,針對於未知環境探索採用SLAM技術,而自主導航的部分則是採用Field D* Lite演算法。原有系統以粒子濾波器為基礎之系統,具有運算時間過長等問題。為解決遭遇到之瓶頸,採用以擴展式卡爾曼濾波器為基礎之SLAM演算法,建構機器人運動模型,以高斯分佈描述機器人的運動誤差,並運用雷射測距儀,對環境擷取「線」特徵與「角點」特徵 (結合Harris偵測演算法和雷射光束模型),作為定位校正的量測依據。於路徑規劃部分,採用Field D* Lite演算法,可以針對不斷變動的環境,快速調整路徑。本系統平均每一步定位誤差小於9%,平均運算時間為1.26秒。若於已知地圖情況下,則每部平均運算時間為0.45秒單步運行時間較舊有系統提升速度達4倍。於建構地圖時,平均速度約為5.2公分/秒,於自主導行時,平均速度為7.5公分/秒,整體而言已具備農業移動平台之實用價值。
In this thesis, an autonomous mobile robot applied to agricultural environment is proposed. To explore the unknown field and navigate with map, the system use SLAM algorithm and the auto path planning technique. Instead of particle filter which costs more time at calculate, our research focus on EKF-based SLAM. Depending on the theory of Kalman filter, all the error (including the motion model and measurement model) follow the Gaussian distribution. The system is equipped with the two-dimensional laser range finder to get the data, and extracts line feature and the corner feature (using Harris corner detection and the beam model of laser) from the contour of environment. Field D* lite algorithm is the main part of the navigation system. It can update the path when the map is changed in real time. The error in localization is less than 9 %, and the average running time is 1.26 seconds per step. Excluding the mapping step, the average running time is speed up to 0.45 seconds per step. With the procedure for SLAM, the average speed is 5.2cm/s. With the procedure for auto navigation, the average speed is 7.5cm/s. The result is feasible for agricultural applications.