Observations on SOFIA Observation Scheduling: Search and Inference with Discrete and Continuous Constraints

In previous work we describe the problem of scheduling flights for SOFIA, an airborne observatory. Scheduling in this domain requires solving many Initial Value Problems (IVPs) and Boundary Value Problems (BVPs) to evaluate constraints on the observatory’s ground track and telescope elevation limits. These are costly operations, and become more so when accounting for winds and fuel consumption. In this paper we show how to reduce the number of IVPs and BVPs needed to schedule SOFIA by restricting the set of flight plans we sample. Empirical studies show that the restriction costs us little in terms of the value of the flight plans we can build. The restriction allowed us to reformulate part of the search problem as a zerofinding problem. The result is a simplified planning model and significant savings in computation time.