In this section we introduce the speed of the ant as a component of the controlling program's fitness. Disapointingly again no significant increase in performance was found.
We investigated two means of combining the ant's speed and the amount of food it eats. In the first speed is added to food eaten to yield a single scalar fitness measure. A threshold of unity is first applied. The threshold means very fast ants have no advantage over ants that are fast enough to complete the whole trail within the time limit and speed accounts for less than one food pellet in the fitness measure. As tournament selection is used the exact scaling co-efficent is unimportant.
We define the speed of the ant as the distance along the trail to the furthest pellet it eats divided by the energy it consumed to reach that food pellet (if eats none, its speed is zero). This is linearly scaled by dividing it by the speed required to complete the trail (i.e. 144/600).
The second mechanism was to treat the scaled speed as a second objective. Multi-objective Pareto tournament selection was used to decide which programs were selected for reproduction using the mechanism described in [Langdon1998b]. In multi-object tournaments it is possible that two or more members of the tournament are not dominated by any others. E.g. one eats less than the other but moves faster. Rather than decide at random between them, we conduct a secondary tournaments against a sample of the rest of the population. The winner being the one which is worse than the smallest number of other individuals from the (sample of the) rest of the population. (We treat having exactly the same fitness as being worse). Sample sizes of 25 and 200 were tried. In [Langdon1998b] use of such a comparsion set was found to be very effective, perhaps as it tends to avoid the population converging on a small number of fitness values. Thresholds of 1.0 and 2.0 on the scaled speed where tried. Again, as Table 2 ``Pareto'' shows, no improvement above that obtained with simply limiting the amount of food ahead of the ant was seen.