Two sensors are available for an observation campaign. Both can observe target A, but only sensor 1 can observe B and only sensor 2 can observe C. All three requests matter. The difficulty is deciding not only which observation to make, but which sensor should do it and when.
The targets' visibility windows are short. A and B can be observed from 10:00 to 10:10, and C from 10:10 to 10:20. Each observation takes ten minutes. The requests carry illustrative values of 12 for A, 10 for B and 9 for C. Sensor 2 can observe A at 10:00 and still observe C at 10:10, because these are back-to-back windows. Assume zero changeover time in this example.
One choice can block the next
A simple planner sees the highest-value observation, A, and assigns it to the first free instrument, sensor 1. That secures 12 units of value. Sensor 2 cannot observe B, so B is lost. Sensor 2 can later complete C, for a total of 21.
The difficulty is deciding not only which observation to make, but which sensor should do it and when.
But consider another schedule. Give B to sensor 1 at 10:00. Give A to sensor 2 at the same time. Then use sensor 2 for C at 10:10. All three observations are completed, for a total value of 31.
| Observation | Window | Eligible sensor | Illustrative value |
|---|---|---|---|
| A | 10:00–10:10 | 1 or 2 | 12 |
| B | 10:00–10:10 | 1 only | 10 |
| C | 10:10–10:20 | 2 only | 9 |
Nothing about the instruments changed. The better arrangement came from looking ahead and recognizing that a flexible task should not use the resource needed by an inflexible one.
Why space observation adds difficulty
Actual observation scheduling has geometry that a small table cannot capture. Visibility depends on orbital motion, sensor pointing, field of view, operating conditions and the timing of other tasks. Windows may last minutes, and some opportunities cannot be recovered once they pass. A sensor may need preparation or recovery time between observations. Targets may have revisit requirements or different kinds of scientific and operational value.
The small example exposes the central insight: a flexible observation is sometimes most valuable precisely because it can be moved to leave room for another.
These features become decisions, constraints and objectives in the optimization model. An algorithm must compare connected opportunities rather than greedily fill each open slot. The quality of the answer depends on both the model and the search method used to navigate it.
A good scheduler can help operators make more valuable observations with limited capacity. If it is fast enough, it can also respond when a new urgent target appears or an instrument becomes unavailable. Reoptimization can preserve work already under way while reconsidering the opportunities that remain.
The small example exposes the central insight: a flexible observation is sometimes most valuable precisely because it can be moved to leave room for another. That is why deciding what to observe is also a problem of deciding what not to displace.