Conformant Planning with Static Causal Laws and Negation as Failure: Decomposition of the ASP Approach

Abstract

In recent years, we have seen a few attempts to use the Answer Set Programming (ASP) to solve the problem of conformant planning with the support for causal laws and negation as failure. Typically, the whole planning domain is encoded into a large ASP meta-program expressing what holds and which actions should be executed in all the time steps of a resulting plan. Answer sets of this meta-program then represent all the possible plans. Since the complexity of answer set computation is exponential in size of input, we propose a technique of dividing it into several smaller meta-programs - one for every state transition. Actual plan finding can then be outsourced to a graph-search algorithm. This method increases the performance of planning with ASP semantics and allows us to produce considerably longer plans. After discussing the worst-case complexity of both approaches, we introduce the prototype ASP planner GRASP that employs our suggested technique and provide the experimental comparison with an alternative planning system DLV K.