Hands-On Reinforcement Learning for Games

Reasoning on 3D worlds

So, why are 3D worlds so important, or are at least believed to be so? Well, it all has to come down to state interpretation, or what we in DRL like to call state representation. A lot of work is being done on better representation of state for RL and other problems. The theory is that being able to represent just key or converged points of state allow us to simplify the problem dramatically. We have looked at doing just that using various techniques over several chapters. Recall how we discretized the state representation of a continuous observation space into a discrete space using a grid mesh. This technique is how we solved more difficult continuous space problems with the tools we had at the time. Over the course ...