TitleEfficiently explaining deterministic exogenous events in partially observable environments
Publication TypeReport
Year of Publication2012
AuthorsMolineaux, M, Aha, DW, Kuter, U
Document NumberAIC-12-081
InstitutionNaval Research Laboratory, Navy Center for Applied Research in Artificial Intelligence
CityWashington, DC
TypeNCARAI Technical Note
Keywordsexogenous events, explanation, partially observable environments

We consider the problem of continual planning (DesJardins et al. 1999) in hazardous partiallyobservable dynamic environments, where deterministic exogenous events that cannot be directly observed affect the state of the world and no plan can be guaranteed to succeed. In these environments, limited observability makes state transitions
ambiguous and difficult to predict. To resolve this ambiguity, we have developed two versions of DISCOVERHISTORY, an algorithm that understands its environment by abductively explaining changes in state through reference to event models. We provide an analysis of their computational complexity and an empirical comparison of their
performance in terms of execution time and success rate at accomplishing goals. We show that use of explanation generation increases the success rate of a continual planning agent and provide an initial benchmark for efficiency of continual planning with explanation in standard domains.

Refereed DesignationNon-Refereed
Full Text
NRL Publication Release Number: 
exogenous events
partially observable environments