- Development of an abductive AI-planner with resources
AI-planning is the field in AI that is concerned with the search for
plans of actions to obtain some goal state starting from some initial
state. Currently, there exist efficient basic search algorithms to
solve this type of problems. However, a problem with these algorithms
is that they often cannot deal with resources. For example, an
observation satellite must plan a sequence of actions to perform a
number of measurements using different instruments. The construction
of the plan must take into account that these instruments share system
resources such as power, fuel, communication bandwidth, processor
time, etc.. To solve planning problems in this type of contexts
requires the ability of reasoning on real number-valued entities (such
as the amount of energy and time). Current basic planning algorithms
lack this ability and therefore cannot handle this type of problems.
The goal of this system is to implement a flexible planning system
based on an abductive reasoning system, the Asystem
. Abductive reasoning is a form of logical inference suited for
searching 'explanations' for formulas. In the context of planning, the
formula to be explained represents the goal state and the explanation
is nothing else than the plan to reach the goal state. The system
will be developed using the A-system, an abductive reasoning system
suitable to reason with real numbered entities. The thesis will start
with a study of the basic planning techniques and an introduction to
the A-system. Then, the A-system must be extended to a planning
system. The system must be validated by applying it in the context of
a number of experiments and comparing it with existing systems.