Chapter 8: Behavioral Cloning
In this chapter, we are going to train a neural network to control the steering wheel of a car, effectively teaching it how to drive a car! Hopefully, you will be surprised by how simple the core of this task is, thanks to deep learning.
To achieve our goal, we will have to modify one of the examples of the CARLA simulator, first to save the images required to create the dataset, then to use our neural network to drive. Our neural network will be inspired by the architecture of Nvidia DAVE-2, and we will also see how to better visualize where the neural network focuses its attention.
In this chapter, we will cover the following topics:
- Teaching a neural network how to drive with behavioral cloning
- The Nvidia DAVE-2 neural network
- Recording images and the steering wheel from Carla
- Recording three video streams
- Creating the neural network
- Training a neural network for regression
- Visualizing the saliency maps
- Integrating...
