Emergence of Prediction by Reinforcement Learning Using a Recurrent Neural Network

Figure 6

Change in the $Q$-value for “catch” action for three velocities. The timing for the increase in value differs, but the position of the object at that timing is almost the same in all cases.