Making Compromises Among Antagonist Constraints in a Planner

Abstract Process planning consists in producing plans for manufacturing parts. It requires taking into account a great variety of pieces of knowledge. Most of them are not pure constraints, but preferences between which compromises are necessary. This is a situation which is typical of many real-life planning tasks. We have designed a planner which operates by iteratively constraining a loosely constrained initial plan built from the model of the part to be machined. If an inconsistent (i.e. overconstrained) plan is produced, then the planner makes use of a selective backtracking procedure which reasons about the weights attached to the applied constraints for deciding which compromise is ‘best’. Successively applied constraints are drawn from ‘pieces of advice’ provided by an expert-type knowledge base. A system, gari , embedding this planner has been implemented and experimented for generating the machining plans of mechanical parts. Some plans were produced after several hundred constraints were applied, among which more than one hundred compromises were necessary.