A Constraint Formulation for Domain Repair with Ground or Lifted Test Plans

Abstract

Engineering automated planning domains is a meticulous and time-consuming process. In recent years, much attention has been given to techniques assisting domain designers in both writing new domains and ensuring their correctness. This paper addresses the problem of repairing a flawed planning domain to comply with a given example problem and plan. The goal is to suggest a minimal set of repairs, which can either be the removal of a condition or effect, or the addition of a new effect in some of the domain's action schemas. The cases of fully specified (ground) and partially specified (lifted) plans, whose actions' parameters can be variables, are handled separately in previous approaches. We introduce an alternative method that covers both ground and lifted cases in a unified way by repurposing an existing encoding of planning as constraint satisfaction. Despite its simplicity, we show the approach to be competitive with the ground case baseline solver and substantially outperforming the lifted one.