In order to measure the
effectiveness of the three proposed methods to enhance the goals, we ran Grt using them in the logistics
problems of the Aips-00
competition. We selected this domain, since in the other domains of the
competition the goal state is either complete, or near complete, so there is no
difference among the three methods. Figure 8 shows the solution length and time
for the easiest of the logistics problems.
With regard to
solution length, the first method, which considers all the candidate facts as
goal facts, always came up with better plans. As we mentioned in Section 3.2,
this method produces small differences among the estimated distances, so the
search process tends to be breadth-first. However, in most of the cases, the
third method found plans of equal quality. With regard to the solution time,
the last two methods work faster, since they produce greater differences
between the distances.
In Section 3.3 we also presented a method of enriching the domain
representation. As already mentioned, we were motivated by the need to treat
domains like the movie or the elevator. We do not present comparative
performance results between the domain enrichment method and the pure Grt planner for these domains, since
without this technique it is impossible for Grt
to solve the problems. However, it would be interesting to test the efficiency
of this method to other heuristic state space planners.
Figure 8: Results for logistics problems
using different methods to complete the goals.
All = Consider all the candidate facts as goal facts.
Initial = Select the initial state facts.
Greedy = Favor the most promising facts.
Ioannis
Refanidis
14-8-2001