XII is based on the UCPOP algorithm [14]. The algorithm we used for interleaving planning and execution closely follows IPEM, by Ambros-Ingerson and Steel [1]. Our action language borrows both from ADL [13] and UWL [4].
The problem of redundant information gathering was addressed by the SOCRATES planner, discussed in [5]. Like XII, SOCRATES utilizes the UNIX domain as its testbed and interleaves planning with execution. Our advances over SOCRATES include the ability to satisfy universally quantified goals and the machinery for automatically generating LCW effects and for detecting threats to LCW links.
Genesereth and Nourbakhsh [7] share our goal of
avoiding redundant information gathering, but do so using radically
different mechanisms, and in the context of state-space search. They
derive completeness-preserving rules for pruning the search as well as
rules for terminating planning and beginning execution. However, they do
not have notions that correspond to LCW, a database like , or our
threat resolution techniques.
Other researchers have investigated alternative approaches for planning with incomplete information (see [12] for a nice taxonomy). Contingent planners [15][16][17] seek to exhaustively enumerate alternative courses of action; while this strategy is appropriate in critical domains with irreversible actions, the exponential increase in planning time is daunting. Decision theory provides an elegant framework for computing the value of information; however, although work in this direction is promising, many challenges remain [18]. Our approach sacrifices the elegance of a probabilistic framework to achieve a complete implementation able to tackle practical problems.