English   español  
Por favor, use este identificador para citar o enlazar a este item: http://hdl.handle.net/10261/134735
logo share SHARE   Add this article to your Mendeley library MendeleyBASE
Visualizar otros formatos: MARC | Dublin Core | RDF | ORE | MODS | METS | DIDL
Exportar a otros formatos:

Improving cost and probability estimates using interaction

AutorEscudero Martín, Yolanda
Palabras claveAutomated planning
Deterministic planning
Low-cost solutions
Heuristic guides
Fecha de publicación4-may-2016
CitaciónImproving cost and probability estimates using interaction : XIX, 225 p. (2016)
Resumen[EN] Automated Planning is concerned with finding a program of actions that given an initial state reaches a desired goal. This program can be a simple sequence or a more complex partially ordered collection of actions. When action outcomes are deterministic and information is perfect, it is called deterministic planning. Significant advances have been made in recent years in the size and complexity of deterministic problems that can be solved with these planners. However, in real problems, unexpected events may occur, actions may have unexpected outcomes, and the state of the world may not be known with certainty. If a deterministic planner is used to solve such problems, the execution of the plan may fail because the plan does not take into account the possible contingencies. Traditional approaches to planning under uncertainty involve Markov Decision Processes, which generates robust plans, but have high computational overhead. Other approaches to planning under uncertainty use contingency planning, translation into deterministic planning, or determinization and replanning. In recent years, planning-based solutions have been also used to solve goal recognition problems. Goal recognition may be seen as the inverse of planning since it is concerned with inferring an agent’s goals given some or all agent’s performed actions. There are few planning-based approaches for goal recognition, but this paradigm is still in its infancy. The previous paradigms all have one thing in common: the dominant techniques to solve their problems involve heuristic functions to guide the planning search. For this reason, in this thesis, we investigate classical heuristics that deal with action costs and introduce a novel domain-independent heuristic function that computes more accurate estimates of cost and estimates of probability. The approach involves a propagation of cost or probability plus Interaction information through a plan graph. This heuristic guides a classical planner to low-cost solutions, guides a probabilistic planner to high probability of success solutions, and rapidly solves goal recognition problems.
DescripciónTesis Doctoral para la obtención del Título de Grado de Doctor. xix, 255 p. : il., diagr. Fecha de defensa de la Tesis Doctoral: 4 de mayo de 2016. Calificación: Sobresaliente cum laudem
Aparece en las colecciones: (CAR) Tesis
Ficheros en este ítem:
Fichero Descripción Tamaño Formato  
Dissertation_E-Martin_Y_ 4_May_16.pdf1,97 MBAdobe PDFVista previa
Mostrar el registro completo

NOTA: Los ítems de Digital.CSIC están protegidos por copyright, con todos los derechos reservados, a menos que se indique lo contrario.