Planification flexible. Un besoin en intelligence ambiante. Un défi en planification automatique
In Revue d'Intelligence Artificielle (RIA) 29(1). pages 11-46. 2015.
Cyrille Martin, Humbert Fiorino, Gaëlle Calvary
Abstract
In order to be used in Ambiant Intelligence, automated planning has to generate highlevel
action plans involving control structures for execution controlers, which role is to play
these plans with respect to events perceived in the environment. In our approach of "flexible"
planning, environment non determinism is managed by plan control structures. In this paper,
we explore how to express and to generate high-level plans for deterministic planning problems
by defining the operational and denotational semantics of new operators for plan composition.
Our Lambda GraphPlan (LGP) planner incorporates into plans iterations representing nondeterministic
choices among a set of resources subject to the same abstract treatment. LGP is a planning-graph based algorithm that extracts patterns of actions whose scheduling is indifferent
with respect to goal reachability and aggregates them into iterative structures. We show
that LGP can be highly efficient when the solution plans incorporate iterative structures.