Page 30 - Read Online
P. 30
Ernest et al. Complex Eng Syst 2023;3:4 I http://dx.doi.org/10.20517/ces.2022.54 Page 11 of 22
Table 1. Individual Components that made up best chromosome fitness of 389.49.
Fitness Component Value
Marines Alive (x25) 100.0
Total Friendly Health Remaining 292.0
Total Hostile Health Remaining 0.0
Timestep Penalty (-Timestep/100.0) -2.51
Total 389.49
Figure 6. Generational and cumulative fitness values of Genetic Algorithm population.
Figure 7. Frame captures 1 (Left) and 2 (Right) of the mission depicting the evaluation of the best chromosome.
In Figure 7 there are 2 frames of the performance displayed. It is a very active mission, with over 70 actions
per second executed by this particular AI model, equating to around 4,200 actions per minute for the roughly
5-9 seconds this mission will last. Obviously this is quite higher than even expert humans would be able to
maintain, but again following humanistic restrictions is not a focus of this particular study. There is non-
determinism, but the problem typically ends with a few of the marines having been defeated, in this case there
were two defeated marines.
Additional training and and increasing the size of the training portfolio would likely lead to greater perfor-
mance, but for the purposes of this formal verification study, the resultant chromosome is sufficient. It is high