MVILAIN@G.BBN.COM (Marc Vilain) (02/04/88)
BBN Science Development Program AI Seminar Series Lecture SYNTHESIZING PLANS THAT CONTAIN ACTIONS WITH CONTEXT-DEPENDENT EFFECTS Edwin P.D. Pednault AT&T Bell Laboratories Holmdel, New Jersey (!vax135!epdp@UCBVAX.BERKELEY.EDU) BBN Labs 10 Moulton Street 3rd floor large conference room 10:30 am, Tuesday February 9th ********************* * note unusual room * ********************* Conventional domain-independent planning systems have typically excluded actions whose effects depend on the situations in which they occur, largely because of the action representations that are employed. However, some of the more interesting actions in the real world have context-dependent effects. In this talk, I will present a planning technique that specifically addresses such actions. The technique is compatible with conventional methods in that plans are constructed via an incremental process of introducing actions and posting subgoals. The key component of the approach is the idea of a secondary precondition. Whereas primary preconditions define executability conditions of actions, a secondary precondition defines a context in which an action produces a desired effect. By introducing and then achieving the appropriate secondary preconditions as additional subgoals to actions, we ensure that the actions are carried out in contexts conducive to producing the effects we desire. The notion of a secondary preconditions will be defined and analyzed. It will also be shown how secondary preconditions can be derived in a general and domain-independent fashion for actions specified in ADL, a STRIPS-like language suitable for describing context-dependent effects. -------