Branching constraint satisfaction problems : sequential constrained decision making under uncertainty
One of the main characteristics of our world is uncertainty. Making plans for the future is difficult, as we do not know exactly what the future holds. Companies must be flexible, ready to cope with the unpredictable demands that are placed on them. As a result, plans are often either short term, or tend to change soon after they are made. Another feature of the modern world is its pace. Decisions must be made quickly, or events may make them out of date before they can be implemented. In this thesis, we look at decision making problems in the presence of uncertainty about how the problem may develop over time, and in particular where the decisions must be made efficiently. Constraint based reasoning has proven to be a very successful technique for supporting decision making, but to date it has assumed static problems. In this thesis, we will show that constraint based methods can be used to reason about uncertain futures, and we will present a method which incorporates some ideas from decision theory to represent and solve such problems. In particular, we will formulate a class of problems, develop systematic optimisation search techniques, incomplete heuristic methods and compare with existing techniques.