Automated feature extraction for planning state representation

• Oscar Sapena ^[1] ; Eva Onaindia ^[1] ; Eliseo Marzal ^[1]
  1. [1] Universidad Politécnica de Valencia

     Universidad Politécnica de Valencia

     Valencia, España

     
• Localización: Inteligencia artificial: Revista Iberoamericana de Inteligencia Artificial, ISSN-e 1988-3064, ISSN 1137-3601, Vol. 27, Nº. 74, 2024, págs. 227-242
• Idioma: inglés
• DOI: 10.4114/intartif.vol27iss74pp227-242
• Enlaces
  • Texto completo
• Resumen
  • español

    Los m ́etodos de aprendizaje profundo han surgido recientemente como un mecanismo para generarembeddings de estados de planificaci ́on sin la necesidad de predefinir espacios de caracter ́ısticas. En este trabajo,abogamos por un enfoque automatizado, eficiente en costes e interpretable para extraer caracter ́ısticas repre-sentativas de los estados de planificaci ́on a partir de un lenguaje de alto nivel. Presentamos una t ́ecnica que sebasa en los tipos de objetos y permite una generalizaci ́on sobre todo un dominio de planificaci ́on, posibilitandola codificaci ́on de informaci ́on num ́erica del estado y de los objetivos de tareas de planificaci ́on individuales. Larepresentaci ́on propuesta se eval ́ua mediante una tarea de aprendizaje de funciones heur ́ısticas para dominiosespec ́ıficos. Un an ́alisis comparativo con uno de los mejores planificadores secuenciales actuales y con un enfoquereciente basado en aprendizaje autom ́atico demuestra la eficacia de nuestro m ́etodo para mejorar el rendimientode los planificadores.

  • English

    Deep learning methods have recently emerged as a mechanism for generating embeddings of planning states without the need to predefine feature spaces. In this work, we advocate for an automated, cost-effective and interpretable approach to extract representative features of planning states from high-level language. We present a technique that builds up on the objects type and yields a generalization over an entire planning domain, enabling to encode numerical state and goal information of individual planning tasks. The proposed representation is then evaluated in a task for learning heuristic functions for particular domains. A comparative analysis with one of the best current sequential planner and a recent ML-based approach demonstrate the efficacy of our method in improving planner performance.

• Referencias bibliográficas
  • Malik Ghallab, Dana Nau, and Paolo Traverso.Automated Planning: Theory and Practice. Elsevier,2004.doi:10.1016/B978-1-55860-856-6.X5000-5.
  • Blai Bonet and H ́ector Geffner. Planning as heuristic search.Artificial Intelligence, 129(1-2):5–33,2001.doi:10.1016/S0004-3702(01)00108-4.
  • J ̈org Hoffmann and Bernhard Nebel. The FF planning system: Fast plan generation through heuristicsearch.J. Artif. Intell. Res., 14:253–302,...
  • Malte Helmert. A planning heuristic based on causal graph analysis. InInternational Conferenceon Automated Planning and Scheduling, pages...
  • Patrik Haslum, Blai Bonet, and Hector Geffner. New admissible heuristics for domain-independentplanning. InAAAI, pages 1163–1168. The MIT...
  • Silvia Richter, Malte Helmert, and Matthias Westphal. Landmarks revisited. In Dieter Fox andCarla P. Gomes, editors,AAAI Conference...
  • Malte Helmert and Carmel Domshlak. Landmarks, critical paths and abstractions: What’s thedifference anyway? InICAPS. AAAI, 2009.
  • Erez Karpas and Carmel Domshlak. Cost-optimal planning with landmarks. In Craig Boutilier,editor,IJCAI, pages 1728–1733, 2009.
  • Otakar Trunda and Roman Bart ́ak. Deep Learning of Heuristics for Domain-independent Plan-ning. In Ana Paula Rocha, Luc Steels,...
  • William Shen, Felipe W. Trevizan, and Sylvie Thi ́ebaux. Learning Domain-Independent PlanningHeuristics with Hypergraph Networks. InICAPS,...
  • Dillon Z. Chen, Sylvie Thi ́ebaux, and Felipe Trevizan. Learning Domain-Independent Heuristics forGrounded and Lifted Planning. InICAPS,...
  • Patrick Ferber, Malte Helmert, and J ̈org Hoffmann. Neural network heuristics for classical planning:A study of hyperparameter space. InECAI,...
  • Liu Yu, Ryo Kuroiwa, and Alex S. Fukunaga. Learning Search-Space Specific Heuristics Using NeuralNetwork. InICAPS Workshop on Heuristics...
  • Patrick Ferber, Florian Geißer, Felipe W. Trevizan, Malte Helmert, and J ̈org Hoffmann. Neural Net-work Heuristic Functions for Classical...
  • Stefan O’Toole, Miquel Ram ́ırez, Nir Lipovetzky, and Adrian R. Pearce. Sampling from Pre-Imagesto Learn Heuristic Functions for Classical...
  • Christer B ̈ackstr ̈om and Bernhard Nebel. Complexity Results for SAS+ Planning.Comput. Intell.,11:625–656, 1995
  • Rafael Vales Bettker, Pedro Minini, Andr ́e G. Pereira, and Marcus Ritt. Understanding SampleGeneration Strategies for Learning...
  • O. Rivlin, T. Hazan, and E. Karpas. Generalized planning with deep reinforcement learning. InBridging the Gap Between AI Planning and...
  • Simon St ̊ahlberg, Blai Bonet, and Hector Geffner. Learning generalized policies without supervisionusing gnns. InInternational Conference...
  • Clement Gehring, Masataro Asai, Rohan Chitnis, Tom Silver, Leslie Pack Kaelbling, Shirin Sohrabi,and Michael Katz. Reinforcement Learning...
  • Blai Bonet and Hector Geffner. Features, projections, and representation change for generalizedplanning. In J ́erˆome Lang, editor,IJCAI,...