本論文探討差速輪型機器人在世界座標中規劃其折線運動路徑,並應用左右兩輪之伺服馬達的編碼器與擴展式卡爾曼濾波器估算其當下的世界座標值,以進行軌跡追蹤,並以超音波感測器偵測相關靜(動)態障礙物,達成避障的功能。基於建築物的特徵性、省能源的操控性及規劃的方便性,我們以折線組成相關的運動路徑。此外,為了達成快速追蹤及省能量的操作,提出了三步驟的軌跡追蹤模式分別為趨近、微調和慣性導引步驟。差速輪型機器人之操控(或軌跡追蹤)並不像車型輪型機器人,能直接以前輪控制方向及後輪控制前進或後退之速度,是故本論文亦推導其操控方法與運動路徑之關係。由於差速輪型機器人之運動學為非線性,也會受到外界雜訊干擾(例如,地面不平造成輪子與地面摩擦力不足打滑或機構鬆脫),以致造成運動軌跡與已規劃的軌跡不同,有鑑於此,我們將以擴展式卡爾曼濾波器估測差速輪型機器人之位置及方位。根據所估測的姿態(即位置及方位)和規劃的路徑,應用所建議的軌跡追蹤模式及馬達速度的模糊分散滑動控制,達成所設定的任務。本論文亦將採用分散主動式嵌入視覺系統,量測其於影像平面座標的運動軌跡,並以類神經網路轉換至世界座標值,然後與應用擴展式卡爾曼濾波器所估算的世界座標值,進行比較。 From the viewpoint of the constraint of house architecture, the energy consumption of differential mobile robot (DMR), and the ease of trajectory planning, a desired (or planning) trajectory made up by a set of piecewise stretching lines is addressed. Then we design the navigation with the following three modes: approach mode, fine-tune mode, and inertia navigation mode, such that the DMR can fast move to the desired trajectory in the manner of energy saving. For the localization of a DMR, the extended Kalman filter (EKF) is employed to estimate the posture of the DMR. Based on the estimated posture and the planning path, a fuzzy decentralized variable structure velocity control of the left and right wheels of the DMR is employed to accomplish the assigned task. The avoidances of various static obstacles with different shapes and sizes, and on different locations of the desired trajectory, and dynamic obstacle in the perpendicular direction of the desired trajectory are achieved by using seven ultrasonic rangers on the DMR and the designed strategy for obstacle avoidance. Finally, the localization between an EKF and a distributed active embedded vision system are compared by various experiments.