Distributing flexibility to enhance robustness in task scheduling problems

Temporal scheduling problems occur naturally in many diverse application domains such as manufacturing, transportation, health and education. A scheduling problem arises if we have a set of temporal events (or variables) and some constraints on those events, and we have to find a schedule, which is an assignment of values to the variables that satisfies the constraints. The execution of schedules in practice is typically surrounded by uncertainty, so that it makes sense to assign intervals rather than fixed times to events. Such a schedule is hypothesized to be more robust to disruptions, as it leaves room for adapting the assignment of exact times to events, to disturbances occurring during execution. In previous work, we have shown how to efficiently compute an assignment of intervals to the variables in a temporal scheduling problem, that maximizes the sum of the lengths of the intervals. We empirically evaluated whether we can further improve the robustness of such a schedule by changing the distribution of intervals. In the current paper, we investigate in more detail how characteristics of the input instances affect different scheduling methods’ robustness properties. From this investigation, we derive three new methods for designing interval schedules, and show them to provide similar or improved robustness.