On Learning Action Costs from Input Plans

TL;DR

This paper introduces a novel problem of learning action costs from input plans to enable plan ranking, and presents an algorithm called LACFIP^k that effectively learns these costs.

Contribution

The paper proposes the first method to learn action costs from unlabeled plans, extending the focus beyond action dynamics to cost modeling in planning.

Findings

LACFIP^k successfully learns action costs from unlabeled plans.

Theoretical analysis supports the correctness of LACFIP^k.

Empirical results demonstrate the effectiveness of the algorithm.

Abstract

Most of the work on learning action models focus on learning the actions' dynamics from input plans. This allows us to specify the valid plans of a planning task. However, very little work focuses on learning action costs, which in turn allows us to rank the different plans. In this paper we introduce a new problem: that of learning the costs of a set of actions such that a set of input plans are optimal under the resulting planning model. To solve this problem we present $LACFIP^k$, an algorithm to learn action's costs from unlabeled input plans. We provide theoretical and empirical results showing how $LACFIP^k$ can successfully solve this task.