Program synthesis for generalized planning
Year: 2018
Type: dissertation
Abstract: Generalized planning is the problem of finding an algorithm-like solution called generalized plan to multiple planning instances. The two main tasks to perform in generalized planning are synthesizing and validating generalized plans. In this thesis, we represent generalized plans as a planning programs, enhanced with conditional goto conditions, or finite state controllers. Then, we compile generalized planning problems to PDDL such that we can compute programs using off-the-shelf classical planners. Because solutions to generalized planning are similar to algorithms, we can build libraries of previous knowledge and reuse them if necessary using a call stack. This feature extends to planning programs with procedures, hierarchical finite state controllers and allows recursion. Finally, we introduce new application areas for planning, e.g. unsupervised classification of instances or context-free grammar generation, by defining non-deterministic choice functions for planning programs.
Author: Javier Segovia‐Aguas
Institution: Universitat Pompeu Fabra
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2904397582
