Table 2.

Summary of best hyperparameters and fitness values at generation number 50.

Parameter name	Symbol	R	R + P	R + N	R + P + N
CACLA + var beta	β	0.0251*	0.01685	0.0001*	0.0156
Initial variance	var0	5.1	4.4	4.7	4.3
Initial iterations	$⌈{\mathrm{\delta }}_0∕\sqrt{\mathit{va}{r}_0}⌉$	10	10	11	10
Discount factor	γ	0.836	0.887	0.715	0.868
Exploration rate	σ	1.4	1.2	1.3	0.9
Exploration rate decay	κ	1	1	1	1
Learning rate Critic	η	0.0001*	0.0276	0.0601	0.0076
Critic MLP in →h1st	Cin→h1st	35	35	25	30
Critic MLP hsecond →out	Ch2nd→out	10	15	10	15
Learning rate Actor	α	0.0426	0.1151	0.0226	0.0751
Actor MLP in →h1st	Ain→h1st	30	30	30	30
Actor MLP hsecond →out	Ah2nd→out	15	15	10	10
Reward	R	10	10	10	10
Punishment	P	N.A.	−0.1	N.A.	−0.1
Avg. fitness (SD)		6.253625 (±5.480689)	8.939111 (±6.389975)	4.765209 (±5.280052)	8.864413 (±6.943758)
Best fitness		1.504085	1.550623	1.425379	1.377808

The fitness is the total reaching distance, in meters, on the testing set, thus the smaller the better.

N.A., not applicable.

The star (*) indicates values that reached their maximum or minimum allowed value.