Emergence of Prediction by Reinforcement Learning Using a Recurrent Neural Network

Table 1

The agent’s ideal and actual performance after learning for three cases of invisibility area.

	Range of the invisibility area			Ideal
	Random	Nothing	Maximum
Average reward	0.685	0.685	0.681	0.742
Percentage with which the agent gets the reward	99.0	98.4	99.9	100
Relative distance between the agent and object when the agent chooses catch action	0.270	0.260	0.296	0.144