The code for this chapter implements the classical reinforcement learning methodology of training a neural network. This neural network is mathematically similar to the one we introduced in Chapter 10, Applying Machine Learning in Robotics, stacking layers of (hidden) nodes to establish a relationship between the states (the input layer) and the actions (the output layer).
The algorithm we will use for reinforcement learning is called Deep Q-Network (DQN) and was introduced in Chapter 11, Machine Learning with OpenAI Gym in the Running an environment section. In the next section, Setting the training task parameters, you will be given the operational description of states, actions, and rewards that characterize the reinforcement learning problem that we are going to solve with ROS.
Next, we will present the training scenarios, and...