Planification flexible. Un besoin en intelligence ambiante. Un défi en planification automatique Article - 2015

Cyrille Martin, Humbert Fiorino, Gaëlle Calvary

Cyrille Martin, Humbert Fiorino, Gaëlle Calvary, « Planification flexible. Un besoin en intelligence ambiante. Un défi en planification automatique  », Revue d’Intelligence Artificielle (RIA), 2015, pp. 11-46

Résumé

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.

Voir la notice complète sur HAL

Actualités