MINISTRY OF EDUCATION AND TRAINING MINISTRY OF NATIONAL DEFENCE MILITARY TECHNICAL ACADEMY NGUYEN THI LAN ANH DEVELOPING EFFICIENT LOCALIZATION AND MOTION PLANNING SYSTEMS FOR A WHEELED MOBILE ROBOT IN A DYNAMIC ENVIRONMENT DOCTORAL DISSERTATION: CONTROL ENGINEERING AND AUTOMATION HA NOI - 2021 luan an MINISTRY OF EDUCATION AND TRAINING MINISTRY OF NATIONAL DEFENCE MILITARY TECHNICAL ACADEMY NGUYEN THI LAN ANH DEVELOPING EFFICIENT LOCALIZATION AND MOTION PLANNING SYSTEMS FOR A WHEELED MOBILE ROBOT IN A DYNAMIC ENVIRONMENT DOCTORAL DISSERTATION Major: CONTROL ENGINEERING AND AUTOMATION Code: 092520216 SUPERVISOR: Assoc. Pham Trung Dung HA NOI - 2021 luan an ASSURANCE I certify that this dissertation is a research work done by the author under the guidance of the research supervisors. The dissertation has used citation information from many different references, and the ci- tation information is clearly stated. Experimental results presented in the dissertation are completely honest and not published by any other author or work.
Author Nguyen Thi Lan Anh luan an ACKNOWLEDGEMENTS First of all, I would like to express my sincere gratitude to my advisor, Assistant Professor Pham Trung Dung, who has been directly guiding me through the PhD progress. His passionate enthusiasm, unwavering dedication to research, and insightful advice have motivated me to carry out this research. I do appreciate all support and opportunities that he has provided to me. Then, I wish to thank my co-supervisor my co-supervisor, Dr.
Truong Xuan Tung, for his valuable advices on my research. He has given and discussed a lot of new issues with me. Working with Dr. Tung, I have learnt how to do research systematically.
His support have motivated me to overcome all challenges in during my PhD journey. Next, I also would like to thank the leaders and all lecturers of the Fac- ulty of Control Engineering, Military Technical Academy for supporting me with favorable conditions and cheerfully helping me in the study and research process. Finally, I must express my very profound gratitude to my parents, to my husband for unfailing support me and always encouraging, to my daughter, Tran Nguyen Khanh An, and my son, Tran Duc Anh for trying to grow up by themselves. This accomplishment would not have been possible without them.
Author Nguyen Thi Lan Anh luan an CONTENTS Contents. iv List of figures. v List of tables. Mobile robot models.
Mobile robot platforms. Kinematic model of differential-drive robot. Bayesian filters for localization systems. Extended Kalman filter algorithm.
The particle filter algorithm. Typical obstacle avoidance algorithms. The dynamic window approach algorithm. Hybrid reciprocal velocity obstacle model.
Timed elastic band technique. Conclusions of the chapter. SENSOR DATA FUSION-BASED LO- CALIZATION ALGORITHMS. Extended Kalman filter-based localization algorithm.
Construction of EKF-based localization algorithm. Results and discussions. Particle filter-based localization algorithm. Construction of PF-based localization algorithm.
Results and discussions. Remarks and discussions. DEVELOPING EFFICIENT MOTION PLANNING SYSTEMS. Proposed enhanced dynamic window approach algorithm.
Construction of the EDWA algorithm. The EDWA algorithm-based navigation framework. Algorithm validation by simulations and experiments. Proposed proactive timed elastic band algorithm.
Construction of the PTEB algorithm. The PTEB algorithm-based navigation framework. 98 ii luan an 4. Remarks and discussion.
Proposed extended timed elastic band algorithm. Construction of the ETEB algorithm. Proposed integrated navigation system. Completed navigation framework.
Experimental setup and results. Conclusions and discussion. CONCLUSIONS AND FUTURE WORKS 125 5. 131 iii luan an ABBREVIATIONS No.
Abbreviation Meaning 1 IMU Inertial Measurement Unit 2 GPS Global Position System 3 KF Kalman Filter 4 EKF External Kalman Filter 5 PF Particle Filter 6 VO Velocity Obstacle 7 RVO Reciprocal Velocity Obstacle 8 HRVO Hybrid Reciprocal Velocity Obstacle 9 DWA Dynamic Window Approach 10 EDWA Enhance Dynamic Window Approach 11 EB Elastic Band 12 TEB Time Elastic Band 13 PTEB Proactive Time Elastic Band 14 ETEB Extended Time Elastic Band 15 ROS Robot Operating System 16 PCL Point Cloud Library iv luan an LIST OF FIGURES 1.1 A general control scheme for autonomous mobile robots.1 Two mobile robot platforms under the study.2 The global reference frame and the robot reference frame.3 The velocity space of the dynamic window approach model. Vs , Va , Vd are the possible velocities, admissible velocities, and dynamic window, respectively.4 Procedure of the hybrid reciprocal velocity obstacles of a robot and an obstacle.5 TEB trajectory representation with n=3 poses .6 The example of exploration graph (a). The block diagram of parallel trajectory planning of time elastic bands (b).1 The block diagram of the proposed autonomous mobile robot localization systems based on the multiple sensor fusion methods.2 The data flow from sensors into the EKF for robot localization.3 The extended Kalman filter-based mobile robot localiza- tion system.4 The proposed approaches .5 The sinusoidal trajectories of the mobile robot in three approaches.6 The circular trajectories of the mobile robot in three ap- proaches.7 The mean error and mean square error of the robot’s po- sition of three approaches in two simulations.8 The simulation results using PF localization .1 The navigation framework for autonomous mobile robot.2 The example scenario of the dynamic environments in- cluding a mobile robot and two dynamic obstacles.3 The efficient navigation system based on the EDWA algorithm78 4.4 The trajectory of the mobile robot and obstacles in Sce- nario 1 and 2.5 The trajectory of the mobile robot and obstacles in Sce- nario 3 and 4.6 The minimum passing distance along the robot’s trajectory.7 The robot’s velocity along the trajectory of mobile robot.8 (a) The Eddie mobile robot platform equipped with a laser rangefinder and a NVIDIA Xavier Developer Kit; (b) The data flow diagram of the proposed framework.9 The experimental results of four experiments.10 The example scenario of the dynamic social environments including a mobile robot and three dynamic obstacles. The robot is requested to navigate to the given goal while avoiding two crossing obstacles o1 and o2 , and a moving forward obstacle o3.
The curved dashed line is the in- tended optimal trajectory of the mobile robot.11 The flowchart of the proposed proactive TEB algorithm.12 The navigation framework based on the PTEB algorithm.13 Four snapshots at four timestamps of the two experiments in the simulation environment.14 A hallway-like scenario with walls, objects, humans, and goals.15 The simulation results of the two experiments. The first row shows the collision index of the conventional TEB al- gorithm. Whereas, the second row illustrates the collision index of the PTEB technique. 102 vi luan an 4.16 The example scenario including a mobile robot and two dynamic obstacles.
The curved dashed line is the intended optimal trajectory of the mobile robot.17 The proposed extended TEB algorithm .18 Four snapshots at four timestamps of the two experiments in the simulation environment.19 The navigation framework based on the ETEB algorithm .20 The simulation results of the two experiments. The first row shows the collision index of the conventional PTEB algorithm. Whereas, the second row illustrates the colli- sion index of the ETEB technique.21 The block diagram of the mobile robot navigation sys- tem utilize the EKF-based localization algorithm and the ETEB-based obstacle avoidance algorithm.22 The example of the human detection and tracking algorithm.23 The comparison result of localization systems.24 The example result of the A*-based path planing algo- rithm (the green curve).25 The QBot-2e mobile robot platforms and the data flow diagram of the proposed framework.26 Three snapshots at three timestamps of the experiment in the sparse environment. 122 vii luan an LIST OF TABLES 3.1 Mean error for the three approaches .2 Mean square error for the three approaches .3 The mean error of the proposed localization system and existing localization system.1 Parameters set in experiments .2 The average passing velocity of the robot .3 Parameters of the Eddie mobile robot platform .4 Parameters set in experiments .5 Parameters of the QBot 2e mobile robot platform.
121 viii luan an Chapter 1 INTRODUCTION 1. Motivation Mobility is an essential navigation issue for autonomous mobile robots. To allow the mobile robots to navigate safely in a real-world environment, the mobile robots must deal with four typical functional blocks of the navigation system [1], as shown in Fig.1, including: (i) perception – the mobile robots must interpret its sensors to extract meaningful infor- mation; (ii) localization – the mobile robots must determine their posi- tion and orientation in the environment. In other words, it answers the question “Where am I?”; (iii) motion planning – includes path planning techniques and obstacle avoidance methods.
The mobile robots utilize it to decide how to act to achieve its goals; and (iv) motor control – the mobile robots must modulate their motor outputs to achieve the desired trajectory, i. Motion Motor Perception Localization planning control Real – world environment Figure 1.1: A general control scheme for autonomous mobile robots. It has been known that, the success in robot navigation requires the 1 luan an success of the four aforementioned fundamental processes, and to im- prove the performance of the robot’s navigation system, the performance of all processes need to be improved. In recent years, several domestic researches in the field of robotics have been published in recent years, such as the publications of Viet- nam Academy of Science and Technology, Institute of Information Tech- nology, Hanoi University of Science and Technology, Vietnam National University, Le Quy Don Technical University and Ho Chi Minh City Uni- versity of Technology.
The domestic works mainly focus on trajectory tracking systems [2], [3], [4] and [5]. Specifically, the authors propose control laws which enable the mobile robots to follow predefined trajec- tories. In this research, the authors also stated that, using these control laws the systems can overcome uncertainties caused by dynamic param- eter variations and external disturbances. In other research direction, some studies [6] and [7] propose algorithms based on extended Kalman filter (EKF) to improve the localization system for mobile robots in un- known environments.
And a few works [8], [9] develop adaptive control algorithms for tracking moving targets by using image features of the target which get from a camera system. Finally, a little researches intro- duce a trajectory planning method in a static environment with known start point and target point [10]. As a result, the mobile robot navi- gation system, especially localization and motion planning systems has not been focused and adequately researched. Therefore, in this research we only focuses on two interesting systems including localization and motion planning systems, which are the scope 2 luan an of the thesis.
Localization is the problem of estimating a robot’s pose relative to its environment from sensor observations. It has been referred as the most fundamental problem to provide the mobile robot with autonomous competences. The challenges of localization are from the inaccuracy and inadequacies of sensors and effects of noise. Firstly, the errors of the mea- surement model, or sensor noise, due to the structural characteristics, resolution and error tolerance of different types of sensors or dynamic environments, such as light conditions, obstacles.
Clearly, the solution here is to take multiple readings into account or multi-sensor fusion to increase the overall information from inputs. Secondly, the errors also can be caused by systematic errors (deterministic) such as the size of uneven wheels, the distance between two unbalance wheels, and they can be eliminated by proper calibration of the system. However, there are still a number of non-systematic (random) errors that remain, such as slipping on the surface, changes in the contact points of the wheel are uneven [1], leading to uncertainties in position estimation over time. In addition, when the mobile robot navigates in the harsh environmental conditions, the information can be interrupted in a short or long interval of time.
Therefore, the mobile robot might have insufficient information for estimating the pose during its navigation.