The Fully Automated Planners

There were eleven competitors in this category, representing at least four distinct planning paradigms (forward search, model-checking, local search and partial order planning). Fully-automated planners accept the PDDL2.1 specifications of domain, initial state and goal and compute solutions solely on the basis of these specifications. No additional control knowledge or guidance is supplied. Fully-automated planners therefore depend on sophisticated search control heuristics and the efficient storage of alternative search branches. A popular current approach to search control is to make use of variants of the relaxed plan idea originally proposed by McDermott unpop and subsequently exploited by Bonet and Geffner hsp and Hoffmann ff.

IxTeT [Laborie GhallabLaborie Ghallab1995] entered as a fully-automated planner, but in retrospect it might have been better classified as a hand-coded planner. The ability to hand-code domain representations might have alleviated some of the problems that arose in making the competition domains accessible to IxTeT. IxTeT does not currently accept domains or problems in PDDL format, so it was necessary to translate the competition domains into its own representation language. No automatic translator between PDDL and the IxTeT input language yet exists and it is not clear that such a translation can be automated. Furthermore, the plan representation of IxTeT is more general than that insisted upon for use in the competition but this was not an advantage given the need to automate the plan validation process. In fact, several plans produced by IxTeT could not be validated. The combination of these difficulties made it impossible to offer any real insights into the performance of IxTeT on the competition domains. Nevertheless, we were pleased that an attempt was made to enter IxTeT: it is important that the competition should not cause a fracture between the members of the research community who are interested in entering and those who have long-established alternative planning technology that cannot be easily reengineered to meet the assumptions underlying the competition.