A new algorithm for probabilistic planning based on multi-objective optimization

Abstract

With the fast development of AI planning, planning technology has been widely applied to robotics and automated cybernetics. Many researchers pay more and more attention to the uncertainty in AI planning, probabilistic planning is a important branch of uncertainty planning. In realistic domains, probabilistic planning often involves multiple objectives, where it aims to generate optimal set of plans to satisfy all these objectives. To date, most of probabilistic plan algorithms have only focused on single objective formulations that bound one of the objectives by making some unnatural assumptions. In this paper, we focus on the probabilistic planning problem with multiple objectives, and we introduce the multi-objective optimization method into probabilistic planning to define the multi-objective value function, we extend the single objective probabilistic algorithm learning depth-first search (LDFS) to its multi-objective counterpart multi-objective LDFS (MLDFS).We explain our implemented algorithm, the objective function we redefined, make conclusion and discuss our future work based on this framework.