Learning to Satisfy Conjunctive Goals

https://doi.org/10.1016/B978-1-55860-247-2.50020-6Get rights and content

Abstract

In earlier work (Hammond et al., 1988; Hammond, 1989b) we argued for an execution-time method for constructing useful plans for conjunctive goals. In domains complex enough to preclude complete preplanning for active goals, some goals must be satisfied opportunistically, while in the pursuit of other goals. The recognition of such an opportunity can indicate that an optimized version of the conjoined plan may be worth constructing and saving, with one indication of the utility simply being that the goals were active at the same time and the opportunity was encountered. The utility of learning such a conjoined plan is not completely clear, particularly for action systems that are flexible enough to use multiple plans at once and interleave them at run time. In this paper we attempt to refine the answer to the question of when this sort of plan should be learned.

References (0)

Cited by (1)

View full text