A Translation Based Approach to Probabilistic Conformant Planning

Abstract

In conformant probabilistic planning (CPP), we are given a set of actions with stochastic effects, a distribution over initial states, a goal condition, and a value 0 < p ≤ 1. Our task is to find a plan π such that the probability that the goal condition holds following the execution of π in the initial state is at least p. In this paper we focus on the problem of CPP with deterministic actions. Motivated by the success of the translation-based approach of Palacious and Geffner [6], we show how deterministic CPP can be reduced to a metric-planning problem. Given a CPP, our planner generates a metric planning problem that contains additional variables. These variables represent the probability of certain facts. Standard actions are modified to update these values so that this semantics of the value of variables is maintained. An empirical evaluation of our planner, comparing it to the best current CPP solver, Probabilistic-FF, shows that it is a promising approach.