Emergence of Prediction by Reinforcement Learning Using a Recurrent Neural Network

Figure 7

Positions where the agent catches the object when the initial object direction of motion varies (with no invisibility area).