Title Page
Contents
LIST OF ACRONYMS 9
요약 10
CHAPTER 1. INTRODUCTION 11
1.1. Background 11
1.2. Objective 13
1.3. Previous study 14
1.3.1. Autonomous vehicle path finding from a microscopic perspective 14
1.3.2. Conditions for optimal path finding for autonomous vehicles 15
1.3.3. Autonomous vehicle path finding through lane changes 15
1.3.4. Autonomous vehicle path finding based on reinforced learning 16
1.4. Research flow 18
CHAPTER 2. LITERATURE REVIEW 19
2.1. Autonomous vehicles operating guidelines 19
2.1.1. ODD(Operational Design Domain) 19
2.1.2. OEDR(Object and Event Detection and Response) 26
2.2. Deep reinforcement learning model 29
2.2.1. Q-Learning 29
2.2.2. Deep Q-Learning 30
2.3. Car following model 31
2.3.1. Car following theory 31
2.3.2. Krauss model 31
2.4. Lane change model 33
2.4.1. Lane change theory 33
2.4.2. SL2015 model 33
CHAPTER 3. METHODOLOGY 35
3.1. The scope of area 35
3.1.1. Spatial range 35
3.1.2. Time range 35
3.1.3. Content range 37
3.2. Data collection and analysis 37
3.2.1. Signal data 37
3.2.2. Traffic data 39
3.3. Simulation environment 39
3.3.1. Traffic simulator : SUMO 39
3.3.2. Environmental configuration 40
3.4. Simulation based on deep reinforcement learning 41
3.4.1. Environment variables 41
3.4.2. Simulation scenario 42
3.4.3. Deep Q-Learning policy 43
3.4.4. Execution results 45
CHAPTER 4. RESULTS AND DISCUSSION 47
4.1. Experiment result 47
4.1.1. Average speed 47
4.1.2. Average waiting time 48
4.1.3. Average queue length 49
4.2. Discussion 50
CHAPTER 5. CONCLUSIONS 51
REFERENCES 52
Table 1. ODD Category Descriptions 19
Table 2. ODD Checklist : Level 4 Highly Automated TNC 21
Table 3. Level 4 HAV/TNC Response Mapping 27
Table 4. The Main Parameters of the SL2015 Model 34
Table 5. Spatial Coverage Suitable for ODD Checklist 36
Table 6. Signal Time by Direction 38
Table 7. Intersection Signal Phase 40
Table 8. Variables in SUMO Car Following Model 41
Fig. 1. Level of Autonomous Driving 12
Fig. 2. Mobility Innovation Roadmap 12
Fig. 3. Global Autonomous Vehicles Market Outlook 13
Fig. 4. Research Flow Chart 18
Fig. 5. Structure of Q-Learning 29
Fig. 6. Structure of Deep Q-Learning 30
Fig. 7. The Spatial Scope of Research 36
Fig. 8. Traffic Light Position 38
Fig. 9. Aerial Image 39
Fig. 10. Intersection Signal Setting 41
Fig. 11. Simulation Scenario 42
Fig. 12. Learning Curve of Cumulative Reward 46
Fig. 13. Total Loss 46
Fig. 14. Collision Rate 46
Fig. 15. Average Speed(SIM vs DQL+SIM) 47
Fig. 16. Average Waiting Time(SIM vs DQL+SIM) 48
Fig. 17. Average Queue Length(SIM vs DQL+SIM) 49