Research And Implementation Of Multi-Objective Probabilistic Planning

The classical intelligent planning request the initial world knowledge is complete, the effects of actions are definite, but in the real world the information of the initial world is not completely, and it is often indefinite. In order to satisfy the intelligent planning technology to apply in the practice, the research of incomplete information and indefinite effects planning problems has already been studied gradually.Among a large number of research methods, probabilistic methods can be more accurate to describe the uncertainty information, and so it has been widespread concerned in the research, the solving method has been gradually matured. At present it has developed many algorithms to solve probabilistic plans, but these algorithms aimed at solving a simple object optimal plan, they have neglected other goals or make hypothesis on other goals. However, in the real world, the planning problem is very complicated, and generating plans for agent to operate in real world confront many challenges. Most planning problem involves multiple objectives, it need to synthetic consider of all these objectives. We apply the multi-objective optimization method to deal with multiple objectives, redefined the objective function, and generate optimal plans set to satisfy all of objectives. We have developed mRTDP ,mHDP and mLDFS algorithms which extends from the single objective probabilistic plan algorithm frame work to multi-objective algorithm by using multi-objective function, it enable them to deal effectively with multi-objective probabilistic planning problems.Based on the algorithms proposed above, we have adopted C++ to develop a new planer--MOPP, which realizes solving multi-objectives in probabilistic planning problems. The experiment proof the system may achieve the anticipated theory effect. This planner development has added new stand point for probabilistic planning theory, has broaden the application of technology, it make the probabilistic planning more suitable for solving real world problems.