Summary

In this chapter, we went through many interesting topics.

We started by describing DAVE-2, an experiment of Nvidia with the goal to demonstrate that a neural network can learn how to drive on a road, and we decided to replicate the same experiment but on a much smaller scale. First, we collected the image from Carla, taking care of recording not only the main camera but also two additional side cameras, to teach the network how to correct errors.

Then, we created our neural network, copying the architecture of DAVE-2, and we trained it for regression, which requires some changes compared to the other training that we did so far. We learned how to generate saliency maps and get a better understanding of where the neural network is focusing its attention. Then, we integrated with Carla and used the network to self-drive the car!

At the end, we learned how to train a neural network using Python generators, and we discussed how this can be used to achieve more sophisticated...