1. Many senses, one truth

In Old Man’s War the soldiers receive enhanced senses and a brain that has to integrate them. The first challenge is not having better eyesight; it is trusting what the new senses report and knowing when one of them is lying. The desktop’s ADCS has the same problem. It has star trackers, sun sensors, magnetometers, gyros, and GNSS. Each reports a different slice of reality. The filter has to decide what to believe.

2. How sensors disagree

Sensors can disagree for many reasons:

  • A star tracker is blinded by moonlight or Earth albedo.
  • A magnetometer is affected by a newly activated payload or thruster.
  • A gyro has drifted beyond its bias estimate.
  • A sun sensor sees a reflection instead of the sun.
  • GNSS is temporarily unavailable during a maneuver or eclipse.

The filter must detect these disagreements and reduce the weight given to the faulty sensor without throwing away good data.

3. What the Resident wonders

  • Should the ADCS use a single Kalman filter or separate estimators for different sensor combinations?
  • How is a sensor fault declared? By residual threshold, by voting, or by a plausibility check against other sensors?
  • What happens when the primary star tracker fails? Does the platform switch to sun/magnetometer mode, or does it declare safe mode?
  • How much attitude accuracy is lost in degraded sensor configurations?
  • Can the platform autonomously recalibrate gyros using star tracker data, or does that require ground intervention?

4. The autonomy boundary

Sensor fusion is a place where autonomy and ground trust intersect. A fully autonomous ADCS can reconfigure itself faster than ground can respond. But if it reconfigures wrongly, it can lose attitude knowledge before anyone notices. The Resident leans toward autonomous fault detection with ground notification, and autonomous recovery only for safe-mode cases.

5. A note for the test arc

The most important ADCS test may be the fault-injection test: deliberately disable or corrupt one sensor at a time and verify that the platform remains pointed and safe. A filter that works in simulation but fails in the real sensor-noise environment is not a working filter.