Self-healing as a combination of consistency checks and conformant planning problems

We introduce the problem of self healing, in which a system is asked to self diagnose and self repair. The two problems of computing the diagnosis and the repair are often solved separately. We show in this paper how to tie these two tasks together: a planner searches a prospective plan on a sample of the belief state; a diagnoser verifies the applicability of the plan and returns a state of the belief state (added to the sample) in which the plan is not applicable. This decomposition of the self healing process avoids the explicit computation of the belief state. Our experiments demonstrate that it scales much better than the traditional approach.