Resource-conscious AI planning with conjunctions and disjunctions

Abstract

The aim of this work is to develop a resource-conscious Artificial Intelligence (AI) planning system, which allows for nondeterminism in the environment. Such planner has a potential in applications where actions in "real world" are considered. The planning process is based on proof search for a fragment of Linear Logic (LL) [10] sequents using a subset of LL rules. As LL is resourceconscious and has additive disjunction connective (represents nondeterminism), LL sequents are used to describe an application domain, whereby every LL sequent represents pre- and postconditions of a particular action execution. We present an idea to use Petri net reachability tree analysis for finding proofs for propositional LL sequents. Game playing is used to solve LL additive disjunctions. From LL proofs plans are extracted which because of underlying LL properties keep track of resources and handle both deterministic and nondeterministic actions.