Safe Learning and Repairing of Numeric Action Models for Planning - Archive ouverte HAL
Communication Dans Un Congrès Année : 2022

Safe Learning and Repairing of Numeric Action Models for Planning

Argaman Aloni Mordoch
  • Fonction : Auteur
  • PersonId : 1163053
Brendan Juba
  • Fonction : Auteur
  • PersonId : 1163054
Roni Stern
  • Fonction : Auteur
  • PersonId : 1163055

Résumé

Automated planners often require a model of the acting agent's actions, given in some planning domain description language. Yet obtaining such an action model is a notoriously hard task. This task is even harder in mission-critical domains, in which a trial-and-error approach for learning how to act is not an option. In such domains, the action model used to generate plans must be safe, in the sense that plans generated with it must be applicable and achieve their goals. The challenge of learning safe action models for planning has been recently addressed for domains in which states are sufficiently described with Boolean variables. In this work, we go beyond this limitation and propose the Numeric Safe Action Model (NSAM) learning algorithm. NSAM runs in time that is polynomial in the number of observations and, under certain conditions, is guaranteed to return safe action models. Experimental results show that NSAM is able to quickly learn a safe action model that can solve the majority of problems in a given domain. Finally, we describe an application of NSAM to to repair action models that are observed to be incorrect.
Fichier principal
Vignette du fichier
DX_with_Numeric_SAM_6_.pdf (1.07 Mo) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-03773786 , version 1 (09-09-2022)
hal-03773786 , version 2 (22-09-2022)

Identifiants

  • HAL Id : hal-03773786 , version 1

Citer

Argaman Aloni Mordoch, Brendan Juba, Roni Stern. Safe Learning and Repairing of Numeric Action Models for Planning. 33rd International Workshop on Principle of Diagnosis – DX 2022, LAAS-CNRS-ANITI, Sep 2022, Toulouse, France. ⟨hal-03773786v1⟩

Collections

DX2022
81 Consultations
92 Téléchargements

Partager

More