In addition to the HARDNUMERIC Satellite domain, a COMPLEX Satellite domain was also considered. The complexity in this domain arises from the fact that it combines temporal actions, with durations dependent on the parameters of the action, with the management of numerically measured resources (in this case, the data store available for acquired data). The problem has a quality that is similar to a knapsack problem, with data having to be packed into the limited stores of appropriate satellites. This is combined with the temporal optimisation problem, which involves ensuring that the satellites are efficiently deployed, moving between targets and capturing data according to their capabilities, their current aspects and their available store. As can be seen from Figure 3, the planners -- fully-automated and hand-coded -- produced plans of quite widely varying quality (lower values are better, here, since quality is measured by makespan). In general, TLPLAN produced the best quality plans, although LPG also produced some high quality plans for the smaller problems. As can be seen in problems 13, 16, 19 and 20, particularly, the fully-automated planners occasionally produced plans with quality diverging very significantly from optimal (we do not actually know what the optimal values are for these problems, but we can obviously consider the best value produced as an upper bound on the optimal value).
|