![]() The current version of FLAP stops when a solution is found instead of finding more solutions. Then, a search process is terminated when one of the following circumstances occur: – A solution plan is found. Since any search can potentially start two new child processes, it is necessary to control the possible exponential growth in the number of parallel search processes. If this child search is stuck in a plateau again, it repeats the same diversification process and starts its own two parallel searches to speedup the progress towards a solution. If a child search is successful in finding an exit to the plateau of its parent search, it will continue searching for a solution plan. The goal of a child search is not to escape from the plateau, but to find a solution plan from the frontier state of Π best, which is likely to be closer to the goal. One process uses f DT G whilst the other one uses f FF, in order to diversify the search in two different directions. In this case, two new A ∗ child searches are started in parallel from Π best, as it can be observed in Figure 1. We consider the main A ∗ search is stuck in a plateau when Π best is not updated in several iterations. For the main A ∗ search, Π best is initially set to the initial empty plan, Π 0, but, when another plan is found with a strictly better heuristic value, Π best is set to that plan. For any A ∗ search in FLAP, the plan with the best heuristic value reached so far, Π best, is stored. Although f is actually faster to compute than f FF, our current implementation of h DT G is not as ro- bust as h FF (see section Limitations and extensions of FLAP ). The main A ∗ search starts from the initial empty plan, Π 0, by using the f FF evaluation function. ![]() In FLAP, we apply a new strategy for plateau escaping based on the ideas proposed in the aforemen- tioned works. – Other approaches adding a diversity with an application to planning include a restarting procedure combined with local search and random walk. – A different strategy lies in combining/alternating different heuristics (or search parameters) to diversify the search directions. They tackle this issue by adding a diversity to search, which is an ability in simultaneously exploring different parts of the search space to bypass large errors in heuristic functions.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |