Choices made by a planner: identifying them, and improving the way in which they are made
Abstract
This thesis discusses the ways in which choices are made by an AI
planner. A detailed examination is made of the prerequisites for choice
making, and a discussion of how the making of good choices can be automated
is included. For a given planner, the prerequisites for choice
making can be split into two parts: finding the types of choice made during
the planning process, and finding the information most relevant to
the making of each type of choice. Two means of automatically making
"good" choices are described: using general planning policies that have
been supplied by the user, and using learned heuristics. These possibilities
are explored for a non-hierarchical version of Tate's NONLIN.