Table 2.

Hyper-parameter values used for the DNN component of DQN and CTDL in the grid world simulations.

Parameter	Value	Description
$L$	3	Number of layers
$U$	[128, 128, 4]	Number of units
$C$	10,000	Number of steps before updating the target network
$B$	32	Batch size for training
$\lambda$	.00025	Learning rate for RMSProp
$\kappa$	.95	Momentum for RMSProp
$\varphi$	.01	Constant for denominator in RMSProp