Autonomous Driving SW Contest
Abstract
We designed an autonomous driving vehicle by modifying a toy car using Arduino and a PC. In the Autonomous Driving SW Contest, our performance was evaluated based on following missions: 1) Road following, 2) Obstacle avoidance, 3) Traffic light recognition, and 4) Autonomous parking.
Leveraging Python and PyTorch, we developed an effective methodology to successfully complete aforementioned missions and achieved 2nd Place in the competition.
Method
1. Road following
- Preprocessing: We calibrated the camera to capture the road image and applied a Gaussian filter to remove noise.
- Control: We used a Canny edge detector to detect the road boundary and computed the steering angle based on the road gradient.
2. Obstacle avoidance
- Object Detection: We trained a YOLOv5 model to detect obstacles in the front of the car.
- Control: We used the detected bounding box to calculate the steering angle and avoid obstacles.
3. Traffic light recognition
- Object Detection: We trained a YOLOv5 model to detect traffic lights.
- Control: We used the detected bounding box to recognize the traffic light color and control the car accordingly.
4. Autonomous parking
- Parking Slot Detection: We used LiDAR sensors to detect parking slots.
- Control: We designed the parking algorithm to compute the steering angle for autonomous parking.
Results
Our autonomous driving vehicle successfully completed all missions and achieved 2nd place in the competition. Our code provides detailed information about our project.