Built a self driving car using a 1/12th scale model, a Raspberry Pi, a generic web camera, and a Google Edge TPU.
- Created an API for controlling the car manually in order to collect training data through the webcam.
- Created a second program whereby the car would use OpenCV in order to detect lanes and then steer the car based on written logic (mathematical equation to keep a heading between two lines). This program would also collect data to be used to train an ML model.
- Used behavioral cloning method to train and implement NVidia's Dave-2 ML model in Google Colab (Jupyter Notebook) utilizing the collected data in order to make steering predictions based on vision.
- Implemented and trained a second model (MobileNet v2 SSD COCO Quantized model) to run on the Google Edge TPU, utilizing the transfer learning method in order to detect objects such as road signs in real-time.
Watch the video below to see how I did it, and the car working in action.
Technologies used:
Python
TensorFlow
OpenCV
Google Colab
