G. Armano, G. Cherchi, and E. Vargiu (Italy)
Abstraction, Planning, Macro-Operators.
The attempt of dealing with the complexity of planning tasks by resorting to abstraction techniques is a central issue in the field of automated planning. Although the generality of the approach has not been proved always useful on domains selected for benchmarking purposes, in our opinion it will play a central role as soon as the focus will move from artificial to real problems. In this case, it will be crucial to have a tool for automatically generating abstraction hierarchies from a domain description. This paper addresses the problem of how to identify macro operators starting from a ground-level description of a domain, to be used for generating useful abstract-level descriptions. In particular, a preliminary release of a system devised to automatically generate abstraction hierarchies has been implemented. Compared to our previous work, this paper reports a step further, in the direction of fully automatizing the process, from both a conceptual and a pragmatic perspective. Conceptually, we refined the process of macro-operators extraction by dealing with the problem of parameters' unification through the exploitation of domain invariants, which can resolve ambiguities that may arise while performing abstraction. Pragmatically, we implemented a system that -given a description of the domain expressed in PDDL outputs a set of macro-operators to be used as a starting point for defining abstract operators. Experimental results highlight the ability of the system to identify suitable macro-operators, used as starting point for populating the abstract level. Such macro-operators usually represent good alternatives to those extracted by a knowledge engineer after a thorough (and sometimes painful!) domain analysis.
Important Links:
Go Back
