Implement autonomous driving algorithms using 1/10 scale model car (Xycar model D). This project was conducted as the final project of the K-Digital-Training: 자율주행 데브코스 program by Grepp. Click the following demo gif to view in Youtube.
- Team organization - four teammates
| Name | Role |
|---|---|
| Hyunwoo Park | Lane detection, Driving control, Lane changing, Parking |
| Yongjae Lee | Image feature matching, Software architecture design, Cartographer tuning, RViz 3D map visualization, YOLO |
| Hyunji Lee | Cartographer tunining, Lane detection, Rotary driving, Obstacle dodging |
| Hui-dong Hwang | {Stop line, Traffic light, speed bump} detection |
- Milestone
| Time | Description |
|---|---|
| 8.23 | Team building |
| 8.25 - 8.27 | Mission analysis and planning |
| 8.30 - 9.3 | Lane detection, Monitoring interface, Stop line, Traffic light, Cartographer tunning |
| 9.6 - 9.10 | Parking |
| 9.13 - 9.16 | Cartographer tunning, Rotary driving, Obstacle dodging, Image feature matching, Code integration, Code refactoring |
| 9.17 | Final competition |
-
Monitoring interface
-
- Upper half: bird eye view from front usb cam
- Cyan colored dots: left lane, lane center, right Lane
- White lines: extracted lines
- small green dots: lidar points
- Lower half: ultra sonic sensors in rear
- Big green dots: ultra sonic sensor result
-
-
RViz 3D map visualization
-
- Floated scanned 3D model of competition track as a background, upon that, showed
/points2,/tracked_posetopic. - Refer this repository to load 3D model in RViz.
-
-
Stop line detection
-
Traffic light detection
-
Speed bump detection
-
Image feature matching
- The result showed too many invalid matching, thus we chose a different approach to recognize images. -> darknet_ros(YOLOv3)
-
Reference path
- After building appropriate local/global constraint status (.pbstream), we made a reference path with pure localization.
-
Project architecture
- Initial design
