Conditional A.I. Planning for Manufacturing Operation

AI Planning has become a very promising technology for supporting the autonomous or assisted design of control programs that guide the operation of manufacturing systems. However, present AI Planning applications to this field only rely on complete knowledge assumptions and, as a consequence, the plans so obtained can only be seen as an advice for an engineer to develop a "real" (that is, a closed-loop, reactive and robust) control program.

Therefore, from the point of view of planning for real world domains, there is a need for developing planning systems for synthesizing truly realistic plans, that is, plans that incorporate the information supplied by sensors (closed-loop and reactive plans), and plans capable of describing alternative courses of action for dealing with forseeable contingencies (robust plans).

In this talk, we will show that, amongst the different planning approaches for dealing with incomplete knowledge, the conditional planning paradigm fits wellto these real-world needs. Then, by introducing a motivating example (based on a real manufacturing system) we will explain the main characteristics of the conditional planning paradigm. The explanation will be focused on the knowledge and action representation, the semantics, and the algorithm that allow to obtain plans that can be considered as realistic representations of control programs.