We are almost nearing the end of our journey into what artificial general intelligence (AGI) is and how deep reinforcement learning (DRL) can be used to help us get there. While it is still questionable whether DRL is indeed the right path to AGI, it is what appears to be our current best option. However, the reason we are questioning DRL is because of its ability or inability to master diverse 3D spaces or worlds, the same 3D spaces we humans and all animals have mastered but something we find very difficult to train RL agents on. In fact, it is the belief of many an AGI researcher that solving the 3D state-space problem could go a long way to solving true general artificial intelligence. We will look at why that is the case in this chapter.
For this chapter, we are going to look at why 3D worlds pose such a unique problem to DRL agents and the ways we can train them...
