Implementing an Intelligent Agent for Optimal Control using Deep Q-Learning

In the previous chapter, we implemented an intelligent agent that used Q-learning to solve the Mountain Car problem from scratch in about seven minutes on a dual-core laptop CPU. In this chapter, we will implement an advanced version of Q-learning called deep Q-learning, which can be used to solve several discrete control problems that are much more complex than the Mountain Car problem. Discrete control problems are (sequential) decision-making problems in which the action space is discretized into a finite number of values. In the previous chapter, the learning agent used a 2-dimensional state-space vector as the input, which contained the information about the position and velocity of the cart to take optimal control actions. In this chapter, we will see how we can implement a learning agent that takes...