IMACS: A Case Study in Real-World Planning

This article illustrates the complexities of real-world planning and how we can create AI planning systems to address them. We describe the IMACS Project (Interactive Manufacturability Analysis and Critiquing System) from the University of Maryland, College Park. IMACS is an automated designer's aid to evaluate the manufacturability of machined parts and suggest design modi cations to improve manufacturability. Over the course of our e orts on IMACS the manufacturing domain continually challenged us to come up with working solutions that would scale to realistic problems. This paper compares and contrasts IMACS's planning techniques with those used in classical AI planning systems and describes (1) how some of IMACS's planning techniques may be useful for AI planning in general, and (2) what challenges need to be overcome by AI planners so that they can be successfully used in manufacturing process planning. Similarities between AI planning techniques and IMACS planning techniques indicate the large unrealized potential of AI planning techniques in solving real-world manufacturing problems. On the other hand, di erences seem to indicate the need for domain-speci c planning techniques. In particular, our experience suggests that process planning for complex machined parts would not be accomplished very well by populating a general purpose planner with domain-speci c knowledge. Instead, we needed to integrate the domain-speci c knowledge into the planning algorithms themselves.