智慧型自動導引車於動態環境中之自主式導航

Abstract

智慧型自動導引車必須能在動態、非結構化的環境中自主導航。在動態環 境中，能正確 估測障礙物的運動資訊對導航控制非常重要。由於現今的 車上感測器要即時地獲取顯式 （Explicit）的運動資訊有所困難，因此 我們提出使用環境預測器來取得內隱式（Implicit）的障礙物運動資訊。 這個預測器是由類神經網路所構成，並以本文所發展出的相對誤差倒傳（ Relative Error Back-propagation ) 法所訓練。藉由車上多個超音波感 測器與 CCD攝影機，它可以有效預測出動態障礙物的未來動向，提供導航 控制做預防的措施。為了達成同時應付多個動態障礙物，本文以虛擬力的 概念發展出對應前述環境預測器之導航法則。在路徑規畫部份，本文發展 出基於雙向距離轉換法（Bidirectional Distance Transform) 的路徑規 畫器。使用這個快速規畫方法可使路徑能在線上修改，所以在遇到非預期 情況時自動導引車也能顯出平順的運動。本文所發展之格子地圖（Grid- map) 式的環境模型以多種感測 穈T資訊融合而建構，使得自動導引車擁 有學習能力。關於室內環境中的自我定位是融合了來自編碼器、陀螺儀、 CCD 攝影機、和多個超音波感測器的資訊來達成。對於所發展的導航方法 ，論文 中提出了模擬和實驗的結果。 An intelligent mobile robot should be able to navigate itself in dynamic and unstructured environments. In such circumstances, the motion of moving obstacles need to be estimated. Due to the difficulties for present-day on-board sensor system to get explicit motion information in real time, we propose to construct an environment predictor to obtain implicit motion information. Multiple sensors including ultrasonic rangefinders and a CCD camera are used to obtain motion prediction. The predictor is constructed by artificial neural networks, which are trained by a relative-error-backpropagation algorithm developed in this dissertation. To handle multiple moving obstacles, a reactive navigation method is developed to work with the predictor based on virtual force concept. Using the developed bidirectional distance-transform path planner, the planned path can be modified on-line based on a learned world model. A method is proposed for absolute position estimation in an indoor environment. Fused sensor data from encoders, gyroscope, CCD camera, and ultrasonic sensors are used to accomplish self-localization. Simulation and experimental results are presented to verify the proposed methods.