Finding Objects Through Stochastic Shortest Path Problems

This paper presents a novel formulation for the problem of finding objects in a known environment while minimizing the search cost. Our approach consists in formalizing this class of problems as Stochastic Shortest Path (SSP) problems, a decision-theoretic framework for probabilistic environments. The obtained problems are solved by using offthe-shelf domain-independent probabilistic planners. The advantages of this approach includes: (i) a well defined optimization problem in which the probability of finding the object is maximized while minimizing the cost of searching for the object; and (ii) being able to take advantage, without any modifications to our model, of any (future) technique in the field of domain-independent probabilistic planners, such as better algorithms and better heuristics. We also contribute by empirically comparing three probabilistic planners algorithms, namely FF-Replan, UCT and SSiPP, using our proposed class of problems.