AIBPO: Combine the Intrinsic Reward and Auxiliary Task for 3D Strategy Game

Table 1

The network structure.

NN1	Parameter	NN2	Parameter	NN3	Parameter
Conv	Conv (64, 3, 3)	Conv	Conv (128,3,3)	Conv	Conv (64, 3, 3)
Conv	Conv (32, 3, 3)	Conv	Conv (32,3,3)	Conv	Conv (32, 3, 3)
Active	ReLU	Active	BN	Fc	Linear (64, 3), reward prediction
Transform	Flatten	Active	ReLU	Fc	Linear (64, 3), reward prediction
Fc	Linear (288, 256)	Transform	Flatten	LSTM	Hidden state 256, state value
Fc	Linear (256, 256)	Fc	Linear (288, 256)	Fc	Linear (256, 1), state value
Actor	Linear (256, 4), a			LSTM	Hidden state 256, action value
Critic	Linear (256, 1), r			Fc	Linear (256, 1), action value
Critic	Linear (256, 1), r