Self-Learning of a Fuzzy Controlled Robot for Multi-Objective Obstacle Boundary Following and Its Fast Navigation in Unknown Environments

本文提出一種移動機器人在未知環境下快速導航的方法。此快速導航方法是由以下要件組成:障礙物的邊界跟隨(OBF)、目標追尋(TS)、凸點追尋(CPS)以及行為監控器。特別的是，CPS是由本文所提出的新型行為，這種行為主要用於縮短導航的路徑和時間。OBF行為不僅賦予機器人在導航途中不會發生碰撞的能力，同時使機器人隨著邊界繞開障礙物得以繼續執行導航到目的地的功能。OBF行為是透過自我學習的模糊控制器(FC)得以實現。此學習是透過多目標資料驅動架構所完成，並考慮了'機器人與障礙物之間保持恆定的距離'以及'機器人的高速移動'之間的權衡。本文將該學習問題表述為具約束的多目標優化問題，並透過多目標連續蟻群優化演算法（MO-FCACO）找到一組非支配的FC解。在TS行為的部分，本文透過兩組模糊比例-積分-微分(PID)控制器得以實現。這兩組控制器分別控制著機器人的方向以及速度。CPS行為使機器人處於OBF/TS行為時能夠尋找更短的導航路徑，機器人會尋找障礙物上凸出的位置作為子目標並執行追尋的動作以此來縮短導航路徑。本文提出一種新型的行為監控器用於管理未知環境下OBF、TS和CPS行為之間的切換。此新型的行為監控器並加入了避免死循環問題的機制。在真實的機器人導航中，本文透過粒子濾波器融合了編碼器以及紅外線傳感定位器各自的優點實現了真實環境中的定位功能。最後本文提出了模擬和實驗來驗證快速導航方案的可行性和優勢。This thesis proposes a fast navigation scheme of a wheeled robot in unknown environments. The navigation scheme consists of Obstacle Boundary Following (OBF), Target Seeking (TS), Convex Point Seeking (CPS) behaviors, and a behavior supervisor. Especially, the new behavior of CPS is proposed to shorten the navigation route and reduce navigation time. The OBF behavior not only endows the robot a collision-free behavior but also enables the robot to detour round obstacles to reach a target. The OBF behavior is achieved by a self-learning fuzzy controller (FC) learned through a multi-objective data-driven configuration that considers the tradeoff between maintaining an accurate robot-obstacle distance and a high moving speed. This thesis formulates this learning problem as a constrained multi-objective optimization problem and finds a set of non-dominated FC solutions through the Multi-Objective Front-guided Continuous Ant Colony Optimization (MO-FCACO). The TS behavior is achieved by two sets of fuzzy proportional–integral–derivative (PID) controllers that control the orientation and speed of the robot. The CPS behavior enables the robot to find a shorter path while executing the OBF/TS behavior by moving toward a new sub-goal found at the convex point in an obstacle. A new behavior supervisor that manages the switching among the OBF, TS, and CPS behaviors in unknown environments is proposed with the additional mechanism of avoiding the dead-cycle problem. In the navigation of a real robot, robot localization through the fusion of encoders and an infrared localization senor using a particle filter is proposed. Finally, this thesis presents simulations and experiments to verify the feasibility and advantage of the fast navigation scheme摘要 i Abstract ii List of figures iv List of Tables ix Chapter 1 Introduction 1 Chapter 2 Multi-objective Fuzzy Control of an Obstacle Boundary Following Robot 7 2.1 Fuzzy Controller 7 2.2 Wheeled Mobile Robot and Obstacle Boundary Following Learning Configuration 8 2.2.1 Mobile Robot Description 9 2.2.2 Learning Configuration and Multi-objective Functions 11 2.3 Multi-objective Front-guided Continuous Ant Colony Optimization (MO-FCACO) 16 Chapter 3 Target Seeking and Control 22 3.1 Orientation control 22 3.2 Speed control 24 Chapter 4 Fast Navigation and Behavior Switching of A Robot in Unknown 26 4.1 Convex Point Seeking (CPS) Behavior 26 4.1.1 CPS in Target Seeking 26 4.1.2 CPS in Obstacle Boundary Following 31 4.2 Behavior Supervisor 34 Chapter 5 Robot Localization 39 Chapter 6 Simulations 43 6.1 Simulation results of the OBF control 43 6.2 Simulation results of the TS control 54 6.3 Simulation results of the fast navigation scheme 56 Chapter 7 Experiments 60 Chapter 8 Conclusion 72 References 7