Domain-level planner

The domain_planner directory contains the source files that are used for domain-level planning. These are:

• build_plan_tree.pl contains the code that constructs the game tree. There is a procedure build_plan_tree that is called recursively to obtain the tree

• incr_plan_rec.pl The incremental procedure is used to apply plan rules to obtain a plan structure. It is called by build_plan_tree procedure to obtain a set of alternatives for a choice node in the game tree. It is a recursive procedure since it needs to incrementally add an act to the plan structure. Each act is added by the find_choice procedure. incremental checks for preconditions to acts, and passes any that are found back to build_plan_tree, so that a chance node can be inserted before the choice node.

• utility.pl The make_choice procedure is used to evaluate the game tree and return the tree's best play. This returned structure is a tree with chance nodes and choice nodes, where all but the best alternative have been pruned at the choice nodes. make_choice calls the procedure e_u which is used to evaluate a game tree. e_u in turn calls make_choice to find the best play to evaluate.

• belief_revision.pl The belief_revision procedure performs ordinary precondition and effect belief revision by updating the belief model. Due to its relatively slow execution time, dry-land belief revision has not yet been integrated so that the ordinary domain level problems run quickly.

• fuzzyise.pl This file contains utilities for the sampling form of the planner, which was used in the experiments of section 4.5. It randomly chooses actual belief states according to the distribution estimated by the belief model.

• dialogue_manager.pl This is a rudimentary dialogue manager, that has a control loop to choose the system's act, perform dry-land and ordinary belief revision, update the history list, then read the user's act, perform dry-land and ordinary belief revision, and update the history list. It was only used once, in the experiment on belief model acquisition, where a user had a dialogue by reading canned English text for the system's act, and then performing his response by selecting from a multiple-choice list. This experiment was reported in section 4.7.

• utils.pl This file contains a set of general-purpose utility procedures.