Knowledge Engineering Tools and Techniques for AI Planning by Unknown

Author:Unknown
Language: eng
Format: epub
ISBN: 9783030385613
Publisher: Springer International Publishing

3.6 Search Process Editor

The search process editor allows the user to control the progress of the search. The user can inspect and run the search by different modalities: with “no interruption,” “step by step” and by “multi steps.” For instance, the user can “pause” the search at any time, inspect the current plan and LA-graph and then continue the search step-by-step, i.e., the search progresses only one step and then waits that the user clicks the command button to proceed for the next step. If the user observes that the heuristics of LPG make an incorrect choice when repairing the selected flaw, the search process editor allows her to move the search one step backward, so that she can intervenes and forces the planner to make an alternative decision among a set of alternatives provided by the system. The multi-step modality is very useful to obtain a graphical animation of the search progress. Under this modality, for each search step all graphs provided by the environment are automatically updated and re-displayed after k-milliseconds, where k is the speed of the animation that can be set by the user.

Moreover, the search process editor gives the user a tool for affecting the future search steps of the planning process by modifying the definition of the search neighborhood for every flaw in the current plan. For example, by inspecting the (partial) plan computed so far, the user realizes that, in order to achieve a desired solution plan, some actions should never be removed from the plan under consideration. The search process editor allows her to specify this constraint to the search, and, in this case, the search process editor sends a special message to the planner imposing that in the rest of the search process the removal of these actions will not be part of any search neighborhood.

The search process editor also allows the user to associate a “breakpoint” with a flaw in the plan under construction. When this happens, the editor sends a special message to the planner containing the selected flaw, which modifies the standard behavior of the planner in the following way. Whenever the planning process of LPG selects such a flaw to repair, the process is interrupted; the system presents all possible options for repairing the flaw to the user; and the user choices one of these options repairing the flaw “by hand.”

For each flaw σ repaired by hand and search neighborhood N, the search process editor memorizes the successor action graph (a graph modification) selected by the user. In the successive search steps, if the planning process of LPG attempts to repair flaw σ again, evaluating a search neighborhood similar to N, then the successor action graph selected by the planner is the graph obtained by performing the graph modification previously selected by the user, which could be different from the action graph that the planner would select from the neighborhood according to its heuristic evaluation. Let N curr be the neighborhood for solving the flaw σ under consideration, the similarity between N curr and N is measured by .

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.