Catalogue of Artificial Intelligence Techniques
Aliases: contingency planning
Keywords: conditional, contingency, planning
Author(s): Stuart Blackwood
Conditional planning is a way to deal with uncertainty when planning. Basically you want to plan in advance so that you can deal with any contingencies that arise. This is fairly simple if the environment is fully observable, as there are no unknown states so less potential outcomes. If, however, the environment is not fully observable then the problem becomes less simple. When the environment is fully observable, the planning may involve loops to deal with problems where the solution is of unbounded length. Also, dealing with contingencies, usually with a branching plan for when one arises, some condition is triggered, and a new route of action needs to be taken. There is nothing too complex about conditional planning in this environment, but there still is a large number of methods and techniques to help with conditional planning. For an environment that is not fully observable, not all states are known. Individual states then need to consist of sets of belief states, the set of possible states. It is no longer simple, as the belief space is exponentially larger than the state space. There are many ways to try and overcome this problem, like quantified boolean formulae, backward searching, regression and many more for various different types of problems when using conditional planning. Some focus on the search heuristics, others try to deal with the exponentially large belief space. In this environment, it can be very complex trying to plan conditionally.
- Stuart Norvig and Peter Russel, Artificial Intelligence: A Modern Approach, 2003, pp.433-440.