1. The computer that ran the air plant

Robert A. Heinlein’s The Moon Is a Harsh Mistress returns as the recalled work because the lunar computer Mycroft — Mike — keeps the air plant, the catapults, and the logistics running while humans argue about politics. The desktop will not be sentient, but it will share the problem: a complex facility in a place where ground control cannot always respond in time, and where someone on board has to make routine decisions.

This entry reads about the basics of spacecraft autonomy architectures.

2. What autonomy means for a spacecraft

Autonomy is the capability of a space segment to manage nominal or contingency operations without ground-segment intervention for a given period. The NASA Autonomous Systems Taxonomy splits autonomous functions into two groups:

  • Function-level autonomy: a single subsystem handles its own sensing, planning, and acting. A star tracker autonomously identifies stars. A reaction-wheel controller autonomously manages momentum.
  • System-level autonomy: the spacecraft as a whole coordinates multiple subsystems to pursue goals, recover from faults, or adapt to new conditions.

NASA’s survey of autonomy for space exploration lists demonstrated capabilities across missions: Deep Space 1’s Remote Agent (1997), Earth Observing-1’s Livingstone model-based diagnosis (2005), Mars rovers’ mixed-initiative planning, Astrobee’s free-flyer navigation, and others.

3. Levels of human involvement

Autonomy is not all-or-nothing. The spectrum runs from fully ground-commanded to fully autonomous, with several stable points in between:

  • Ground in the loop: every action is commanded from Earth.
  • Ground on the loop: the spacecraft executes procedures and reports back; humans supervise and can intervene.
  • Ground off the loop: the spacecraft handles routine and contingency operations for long periods without waiting for instructions.

The Deep Space 1 Remote Agent experiment was an early demonstration of ground-off-the-loop planning. It used model-based reasoning, constraint-based planning, and execution monitoring. It could plan activities, detect discrepancies between expected and observed behavior, and replan.

4. ESA’s path: PROBA

ESA’s PROBA-1 was the agency’s first fully autonomous spacecraft. It performed flight-dynamics computations, orbital navigation, and routine operations on board with minimal ground intervention. PROBA-2 extended this with GPS-based navigation, Kalman filtering, and event prediction. The PROBA line shows that small platforms can carry substantial autonomy if the software is designed for it.

5. What this changes

  • Spacecraft autonomy has both function-level and system-level layers.
  • The desktop will need system-level autonomy because cells, attachments, power, thermal, and communications must be coordinated.
  • Ground-off-the-loop operation is feasible and has been demonstrated, but it requires explicit fault management.
  • The next entry will read about fault management and FDIR.