The genetic programming system and the artificial ant problem [Koza1992, pages 147--155,] were set up identically to [Langdon and Poli1998] except the new training technique was used (see Table 1).
The program primitives are the same as before except now food items on the trail are numbered in the order we expect the ant to eat them. Only the first x are placed on the grid initially. As the ant moves and eats them new food pellets are added to the grid. When it eats food pellet n then we ensure all the uneaten food pellets up to pellet n+x are on the grid. 50 (or 100 in the case of x=5) independent runs where carried out with each value of x.
Table 1: Ant Problem