Phase 1: Create an RC car that can be controlled with keyboard controls in order to collect driving data. Every time a movement command [forward, left, right] is issued to the car, the car takes a picture of its current surroundings as well as moves a step according to the input command.
Phase 2: Use TensorFlow to create and train a deep learning neural network model with the training data to predict what direction the car should move toward in a given situation. The training maximizes the accuracy of the model’s predictions with the given training data.
Phase 3: Use the trained model in real time, so the car can drive autonomously.