MAE148-Autonomous Car Project

an Autonomous Driving Project

In this project, my teammates and I built an autonomous car from scratch. Below is a diagram illustrating the car’s construction and appearance:

Through this project, we trained the car to accomplish several tasks, including:

  • Completing autonomous laps using the DonkeyCar KerasCNN model (input: RGB images; output: throttle and steering angle)
  • Completing autonomous laps using GPS
  • Performing autonomous lane-following laps using OpenCV (cv2)

Final proposed task: automatically detecting and reaching a broken-down car

  • Setup: We used a surveillance camera to monitor the road. The car was equipped with both lidar and a camera. When the surveillance camera detected a blinking hazard light, it signaled the car to search for the broken-down vehicle and park behind it to provide assistance.

  • Solutions:

    • The surveillance camera used a blinking light detection model developed with OpenCV (cv2). When a hazard light was detected, a signal was sent to the car via a ROS2 topic.
    • The car used an object detection model served through Roboflow to identify the broken-down car and employed a PID control algorithm to approach it.
    • Lidar was used to determine the distance to the target. A basic sensor fusion technique was implemented to accurately map the object detected by the camera to the lidar output.

Final project demo video:

Resources

References

  • MAE 148 course - Introduction to Autonomous Vehicle course.

  • DonkeyCar - An open source Self Driving Car Platform.

  • Roboflow - A platform to build and deploy computer vision applications.