- 學位論文
追循障礙斥力場谷線之雙眼導航自主駕駛系統
Stereovision-based Autonomous Navigation System by Tracking The Potential Field Valley
摘要
本研究的目的是發展立體視覺導航的自主駕駛系統，並將其應用於行動機器人，此自主駕駛系統分為雙眼立體視覺、行進規劃、運動控制等三個次系統。雙眼立體視覺次系統主要是模擬人類雙眼視覺來偵測環境、取得資訊，其方法是利用影像區域的立體匹配來偵測像差並建立像差影像，接著將像差影像轉換成環境的立體空間資訊；由於建立整幅像差影像非常耗時，為節省演算時間，本研究採用先粹取影像邊緣，再求取邊緣的像差及其三維資訊。行進規劃次系統是利用雙眼立體視覺所得到的環境資訊做機器人的行進規劃，由於雙眼立體視覺的每次取像僅能得到局部的環境資訊，因此，行進規劃也是利用此局部性的資訊來規劃路線，在機器人與目的地之間建立一道吸引力場，在機器人與障礙物之間則建立一道排斥力場，再將兩種力場結合成一道人造能場，能場的谷底線即是所欲的行進路線。由於能場演算法可能發生局部性最小值陷阱和其他不利於機器人行進的情況，因此，本研究設計了四種行進策略來解決。運動控制次系統用適應性類神經模糊推論系統作為運動控制器的基礎，此控制器可以預設專家的經驗所產生的模糊控制法則，也可以經由學習過程來改變其參數，因此具有適應性的功能，適合用於行動機器人的動態控制，最後，本研究製作完成一套自主行動機器人實驗系統並施行多項實驗驗證，證實所發展的雙眼導航自主駕駛系統確實可以成功運作。
並列摘要
The goal of this research is to develop a stereovision guided autonomous navigation system and apply it to the mobile robot control. The autonomous navigation system is composed of a binocular stereovision sub-system, a motion planning sub-system, and a motion control sub-system. The binocular stereovision mimics human visual perception to obtain the 3D information about the environment. The area-based stereo matching algorithm is used to construct the disparity image. The disparity image is then transformed to obtain the 3D information. Since constructing a full frame of the disparity image is time consuming, only the image edges are extracted for constructing the sparse disparity image and the 3D information. The motion planning sub-system plans a local path by according to the local 3D information obtained by a snap shoot of the stereovision sub-system. The potential field algorithm is used to build an artificial potential field for the motion planning. The attractive filed is modeled to emit an attractive force from the goal configuration to the robot. The repulsive field is modeled to emit a repulsive force from the obstacles. The potential field is the sum of the attractive and repulsive field, and the valley shows a desired path. Four motion strategies are designed to deal with the situations of the local minimum traps and other unfavorable conditions. Adaptive Neuro-Fuzzy Inference System (ANFIS) is adopted as the base of the motion control sub-system. The controller allows installing fuzzy rule base obtained by according to expert’s experience, and adapting the parameters through an on-line learning process. This excellent ability makes it suitable control the complex nonlinear behavior of the mobile robot. An experimental system of autonomous mobile robot has been built and several experiments for verifying the proposed designs have been conducted. Experimental results show that the stereovision based autonomous navigation system is feasible and successful.