next up previous
Next: Retrieval Strategy Up: An Empirical Evaluation of Previous: An Empirical Evaluation of


Experiments were run on problems drawn from two domains. The first was the artificial domain, $ \theta _2 D^m S^1$, originally described in  [2] and shown in Figure 8. Testing was done on problems which were randomly generated from this domain with the restriction that they always contain the goal $g_\alpha$. The Logistics Transportation domain of [40] was adopted for the second set of experiments. Eight packages and one airplane were randomly distributed over four cities. Problem goals represented the task of getting one or more packages to a single destination airport[*]. The FLY operator was augmented with a delete condition which prevented planes from visiting the same airport more than once. This meant that replay failed if there was an extra package to be transported which was off the previous route taken by the plane.