Heuristic planning with resources
Ioannis Refanidis, Ioannis P. Vlahavas · European Conference on Artificial Intelligence · 2000
This paper presents GRT-R, an enhanced version of the GRT planner capable of explicitly handling resources. GRT is a domain independent heuristic STRIPS planner, which works in the space of the states. The heuristic computes off-line, in a preprocessing phase, estimates for the distances between the domain's facts and the goals. These estimates are utilized during the search process, in order to obtain values for the distances between the intermediate states and the goals. We propose an explicit representation of resources in a numerical way. There are two kinds of resources: the consumable and the allocable ones. In the pre-processing phase, GRT-R assigns to the domain's facts vectors of costs. The first cost is an estimate of the distance between a fact and the goals, while the remaining costs estimate the amount of consumable resources needed to achieve that fact. GRT-R assigns each fact with a set of such vectors, each one of them corresponding to a different way of achieving the fact. During the search process, GRT-R assigns similarly each intermediate state with such a vector, based on the vectors of the state's facts, with the intention to minimize the distance between the state and the goals, without exceeding the available resources. Allocable resources are taken into account only while searching, in order to preserve the validity of the states. Performance results show that GRT-R copes well in domains that embody resources.