Implementing Constraint Relaxation over Finite Domains Using Assumption-Based Truth Maintenance Systems

Many real-life Constraint Satisfaction Problems are over constrained. In order to provide some kind of solution for such problems, this paper proposes a constraint relaxation mechanism fully integrated with the constraint solver. Such a constraint relaxation system must be able to perform two fundamental tasks: identification of constraints to relax and efficient constraint suppression. Assumption-based Truth Maintenance Systems propose a uniform framework to tackle those requirements. The main idea of our proposal is to use the ATMS to record and efficiently use all the information provided by the constraint solver while checking consistency. We detail the use of ATMS in our particular scheme and enlight their efficiency by comparing them with existing algorithms or systems (Menezes' IHCS and Bessiere's DnAC4).