林巍聳臺灣大學：電機工程學研究所楊秉杰Yang, Ping-ChiehPing-ChiehYang2007-11-262018-07-062007-11-262018-07-062006http://ntur.lib.ntu.edu.tw//handle/246246/52977本研究旨在發展視覺導航行動機器人的追跡控制系統，此追跡控制系統分為自評自調速度控制和姿態控制兩個部份。自評自調速度控制使行動機器人可以自動適應各種路面，本研究採用雙啟發規劃法推導其學習驗算法。姿態控制器由類神經網路所構成，控制法則是經由學習程序自動建立，無需人工進行控制系統分析與設計。姿態控制類神經網路所學習的對象是行動機器人的逆向駕駛模型，所以用各種不同速度連續駕駛行動機器人，其速度和位移的記錄就可以做為訓練樣本。視覺導航系統可以看到遠方，因此可以提供未來即將行進的軌跡，這種預測未來的資訊可以在追跡控制系統內加以運用，讓行動機器人可以為即將發生的變化預作準備。路線規劃器和姿態控制類神經網路的學習機制都有將這種高階控制理念列入設計。電腦模擬各種學習和控制狀況的結果顯示整個追跡控制系統的設計具實用性。The trajectory tracking controller of a wheeled mobile robot (WMR) is designed to compose an adaptive critic velocity controller and a neural network based posture controller. The learning algorithm of the adaptive critic velocity controller is derived by using dual heuristic programming (DHP) method. This adaptive critic design enables the WMR system to comply with variant road conditions. The posture controller design uses MLP neural network to learn the inverse drive model of WMR. The training patterns are obtained by continuously driving WMR with a variety of velocities. Vision information about a desired trajectory in the near future is utilized in both path planner and posture controller. This information can prepare the WMR system for changes in the trajectory in advance. Extensive simulation studies have verified the feasibility of the proposed design.摘要 i ABSTRACT iii Chapter 1 1 Introduction 1 1.1 Background of this research 1 1.2 Motivation and Contribution 3 1.3 Organization of This Thesis 4 Chapter 2 5 Design Methods of Adaptive Critic Neural Control System 5 2.1 Introduction 5 2.2 Adaptive critic neural control with heuristic dynamic programming 9 2.3 Adaptive critic neural control with dual heuristic programming 13 2.4 Adaptive critic neural control with globalized dual heuristic programming 17 2.5 Summary 20 Chapter 3 23 Design of Adaptive Critic Trajectory Tracking Control System 23 3.1 Architecture of the trajectory tracking control system 23 3.2 Adaptive critic design of the velocity controller 24 3.2.1 Architecture of adaptive critic velocity controller 25 3.2.2 The utility function 25 3.2.3 The action network 26 3.2.4 The critic network and the verification network 27 3.2.5 The plant model 29 3.2.6 Training algorithm of the adaptive critic velocity controller 29 3.3 Design of the posture controller 31 3.3.1 Learning the inverse drive model by neural network 31 3.3.2 Learning of neural network 34 3.4 Path planning 36 3.4.1 Path planner 36 3.4.2 Position planner 37 Chapter 4 41 Adaptive Critic Trajectory Tracking Control of Wheeled Mobile Robot 41 4.1 Model of wheeled mobile robot (WMR) 41 4.2 Adaptive critic trajectory tracking control of WMR 44 4.3 Simulation studies of adaptive critic WMR control 46 4.3.1 Performance of adaptive critic velocity controller 46 4.3.2 Performance of neural network based inverse drive model 57 4.3.3 Performance of trajectory tracking control 71 Chapter 5 83 Conclusion 83 Appendix A 85 Parameters of the trained control system 85 A.1 Parameters of the trained posture controller 85 A.1.1 Parameters of the trained linear inverse drive model neural network in section 3.3.1 85 A.1.2 Parameters of the trained angular inverse drive model neural network in section 3.3.1 87 A.2 Parameters of the trained velocity controller 89 A.2.1 Parameters of the trained action network in section 3.2.3 89 A.2.2 Parameters of the trained critic and verification networks in section 3.2.4 89 References 91939901 bytesapplication/pdfen-US自評自調法追跡控制機器人類神經網路Adaptive critic designtracking controlmobile robotneural networkthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/52977/1/ntu-95-R93921078-1.pdf