Space Operations as a Guide for a Real-World Scheduling Competition Eduardo Romero and Marcelo Oglietti CONAE - Argentine National Space Agency Paseo Col´ on 751, (1063) Buenos Aires Argentina {eromero, marcelo.oglietti}@conae.gov.ar Abstract The ultimate objective of the scheduling competition is to drive us developing frameworks that tackle real-life problems better. Whatever form we gave to this competition, bench- mark problems will be the essential tools for identifying the best algorithms and frameworks. So, the future of the compe- tition will be greatly influenced by the problems we choose. Oversimplified ones will produce a distortion from the com- petition’s final objective. In this paper, we present three real-life scheduling problems from space mission operations. We extracted their core features to build a rich benchmark scheduling problem to be used in the competition. Introduction In view of a Scheduling competition, several issues need to be debated. Two important and related issues that must be addressed are: First, what features should a scheduling sys- tem have; and second, what encompass the core character- istics of what we call scheduling. In this article we try to answer these issues by means of describing three real-world problems emerged naturally from the normal operations of satellite missions. Planning and scheduling space operations are critical tasks that consume many resources not only directly, but also indirectly. A suboptimal scheduling implies the under- exploitation of many resources of a space mission and, as a result, a significant increase of the final cost of the obtained products. The need of planning and scheduling tools for space mis- sion is a well-known fact. Accordingly, there is a lot of lit- erature about the successful use of planning and schedul- ing techniques in these domains and since 1997, and every two years, it takes place the International Workshop on Plan- ning and Scheduling for Space. Two recent examples are the scheduling of the Hubble Space Telescope activities (Fer- dous & Giuliano 2006); and scheduling the services of ESA ground stations network (Damiani et al. 2006). This shows an important role played by scheduling tools in cost effective space missions, and how the richness of space-mission operation scenarios makes them perfect can- didates to be benchmark problems for a scheduling compe- tition. Copyright c  2007, American Association for Artificial Intelli- gence (www.aaai.org). All rights reserved. The scheduling problems we present here share some core characteristics that differentiate them from classical scheduling approaches. These differences can be summa- rized as follows: • If we draw a parallel between the problems presented here and classical scheduling problems, we see that our prob- lems are in some sense more constrained: we only have a few, highly tested and documented ways of carry out the tasks. This turns the temporal networks associated with our problems locally very constrained. • Since preference and priority levels are spread all over our examples, we need suitable quality measures for the solu- tions. In classical scheduling we must optimize the used time to process a set of jobs or tasks. In our problems we also have to optimize over a family of quality and priority measures that increase the complexity of the problems. • We need the modeling of a special kind of unary resources that might be configured in different ways. In all the prob- lems we present here, the resources are basically unary. Noteworthy, the use of these resources implies the set of different parameters for their configurations. This infor- mation must be included not only because it is necessary for operations, but because it is used for constraint propa- gation. For example, in order to know whether two units can work in parallel or not according their configurations. Besides, some units can be configured in different ways for the same task. The actually used configuration for a task depends on the configuration of the other units in- volved. • Changes in the schedule are continuously requested in the problems presented here, sometimes with no regularity, thus, we have to deal constantly with re-scheduling and a variable scheduling horizon. Besides, when a schedule has been set we send a notification of this to our clients, and for this reason re-schedule has a cost. Therefore, when re-scheduling, we have to modify as few as possible the previous schedule. In the next section we describe the context in which these problems arise.