CRIKEY! ― It's co-ordination in temporal planning

Abstract

Temporal planning contains aspects of both planning and scheduling. Many temporal planners assume a loose coupling between these two sub-problems in the form of "blackbox" durative actions, where the state of the world is not known during the action's execution. This reduces the size of the search space and so simplifies the temporal planning problem, restricting what can be modelled. In particular, the simplification makes it impossible to model co-ordination, where actions must be executed concurrently to achieve a desired effect. Coordination results from logical and temporal constraints that must both be met, and for this reason, the planner and scheduler must communicate in order to find a valid temporal plan. This communication effectively increases the size of the search space, so must be done intelligently and as little as possible to limit this increase. This thesis contributes a comprehensive analysis of where temporal constraints appear in temporal planning problems. It introduces the notions of minimum and maximum temporal constraints, and with these isolates where the planning and scheduling are coupled together tightly, in the form of co-ordination, it characterises this with the new concepts of envelopes and contents. A new temporal planner written, called СRIKЕҮ, uses this theory to solve temporal problems involving co-ordination that other planners are unable to solve. However, it does this intelligently, using this theory to minimise the communication between the sub-solvers, and so does not expand the search space unnecessarily. The novel search space that CRIKEY uses docs not specify the timings of future events and this allows for the handling of duration inequalities, which again, few other temporal planners are able to solve. Results presented show СRIKЕҮ to be a competitive planner, whilst not making the same simplifying assumptions that other temporal planners make as to the nature of temporal planning problems.