Applying statistical inference to planning under uncertainty
Nathaniel G. Martin · 1993
Because planning involves reasoning about the future, uncertainty permeates it. Some recent planners represent uncertainty about actions and observation by probability. Unfortunately, unless experts are available to quantify this uncertainty, it is difficult to assess them. This dissertation explores techniques for representing knowledge so that a planning system can use its experience to reason about probability in uncertain situations. To support this type of knowledge representation, the dissertation develops a formal event-based language in which the planner's probabilities are calculated from the binomial random variable generated by the ratio of one type of event to another. Inferences about the long-run ratio of these events can be made using statistics. Inferences about the validity of the approximations provided by the statistics allow the planner to avoid making choices that are only weakly supported by the planner's evidence. To test the techniques investigated in the development of the formal language, a planning language, based on the formal language, was developed. The complexity of reasoning about probability makes generating plans that take full account of uncertainty impossible. As an alternative to a planner that uses the planning language, a planning assistant was designed and implemented. This planning assistant performs three services for a logic-based planner. It advises the planner about appropriate actions, executes plans generated by the planner, and monitors the plans it executes.