Title Page
Contents
ABSTRACT 13
Ⅰ. INTRODUCTION 15
1. Research Background 15
2. Research Overview 19
Ⅱ. ROBOT VEHICLE CONFIGURATION 21
1. Robot Vehicle Model 21
1.1. Scale and Body type 21
1.2. Motors 22
1.3. Battery 23
2. Electronics 24
2.1. Embedded System 24
2.2. Servo Motor and ESC driver 26
2.3. Camera 27
2.4. Encoder 28
3. Robot Vehicle Integration 29
Ⅲ. AI-POWERED HYBRID CONTROL ALGORITHM 30
1. Behavior Cloning Algorithm 30
1.1. Dataset Acquisition and Preprocessing 30
1.2. CNN-based Behavior Cloning Algorithm 32
1.3. Evaluation of CNN 33
2. Rule-based Control Algorithm 35
2.1. Vision-based PID Control Algorithm 35
2.2. Verification of Heading Angular Error Calculation 37
2.3. Gain Tuning for PID Control 39
3. AI-powered Hybrid Control Algorithm 41
Ⅳ. VEHICLE TEST RESULTS 45
1. Test Environment 45
2. Cornering and Straight course 46
2.1. Counterclockwise 46
2.2. Clockwise 49
3. Unseen Environment 56
Ⅴ. DISCUSSION 57
Ⅵ. CONCLUSION 59
Ⅶ. References 60
ABSTRACT IN KOREAN 65
〈Table 1〉 Actuators specification 25
〈Table 2〉 Camera specification 27
〈Table 3〉 Range of HSV color model 31
〈Table 4〉 Training results of CNN 34
〈Table 5〉 Driving results of each algorithm: Counterclockwise 47
〈Table 6〉 Driving results of each algorithm: Clockwise 50
〈Table 7〉 Driving results of PID control algorithm in straight course 53
〈Table 8〉 Driving results of PID control algorithm in cornering course 54
〈Table 9〉 Driving results with various steering input ratios: Unseen environment 56
Figure 1. Robot vehicle using RC car model. 21
Figure 2. Motors and electronic speed controller. 22
Figure 3. Nickel metal hydride battery. 23
Figure 4. Embedded systems. 25
Figure 5. PCA9685. 26
Figure 6. RaspberryPi camera V2. 27
Figure 7. Encoder. 28
Figure 8. Robot vehicle overview. 29
Figure 9. Hardware connection of the robot vehicle. 29
Figure 10. Architecture of overall CNN. 32
Figure 11. Training results of CNN. 33
Figure 12. Method of finding a heading angular error. 36
Figure 13. Verification of heading angular error calculation method in a straight course. 38
Figure 14. Verification of heading angular error calculation method in a cornering course. 38
Figure 15. Simulink diagram of PID control. 39
Figure 16. Simulation results of PID controller with various gains. 40
Figure 17. Architecture of AI-powered hybrid control algorithm. 42
Figure 18. Block diagram of AI-powered hybrid control algorithm. 43
Figure 19. Steering control input ratio of AI-powered hybrid control algorithm. 44
Figure 20. Test environments. 45
Figure 21. Counterclockwise vehicle test results. 48
Figure 22. Clockwise vehicle test results. 51
Figure 23. PID control algorithm vehicle test results in straight. 53
Figure 24. PID control algorithm vehicle test results in cornering. 54
Figure 25. Hybrid control algorithm vehicle test results with different PID control algorithm. 55