Representation and automatic synthesis of reaction plans

Rather than committing to a particular future course of events, reaction plans prescribe reactions to be used if various situations should occur. In this respect reaction plans are closely related to both production systems and process control programs. The "Universal Plans" representation is an extreme member of the class of reaction plans, making no assumptions whatever about the future. This Dissertation introduces Universal Plans and shows how they can be automatically constructed from a declarative description of the effects of individual actions.Even before there has been any planning, the Universal Plans executor will be able--in most situations, and with searching--to find something to do next. Of course, the reaction suggested by such an "unplanned plan" might be both too late and seriously wrong. Universal Plans at this stage of inexperience might be regarded as "muddling along". The planner solves search problems and avoids pitfalls for the executor by adding new rules to the rulebase initially received as the domain description. The planner's additions are best regarded as advice for the plan executor.Because Universal Plans are highly conditional, almost all the effects of any plan are conditional. Further, the conditional effects of plans can be conveniently controlled by limiting the circumstances in which a plan is allowed to execute. This possibility gives rise to a new planning technique I call "confinement" that is capable of solving even mutual goal conflicts without interleaving the goal trees of the interacting subplans. This has several important consequences. (1) Subplans can now be reused without modification: a plan that achieves P $\wedge$ Q utilizes in a straightforward way the subplans that achieve P and Q separately. Thus the Universal Plans representation is also a programming language with an effect-preserving composition construct. (2) Because Universal Plans are reactive rather than predictive, the planner is liberated from part of the classical frame problem (the part recently renamed the "persistence" or "inertia" problem).U of I OnlyETDs are only available to UIUC Users without author permissio