Count-Based Exploration via Embedded State Space for Deep Reinforcement Learning

Figure 1

F1-race game screen. Red box line: state features that do not affect the choice of actions. Yellow box line: state features that affect the choice of actions.