The person sets goals while the machine executes details
Aliases: goal-level control · management by exception · human supervisory control
What it is
Supervisory control has a person set goals, constraints, and priorities while the robot plans and executes continuous detail; the person monitors and handles exceptions. It is neither teleoperation nor abandonment: human work remains in goal validity, boundary judgement, and intervention.
Why it happens
Machines handle frequent stable control while people integrate context and informal goals. Hierarchy reduces operation and communication but removes exposure to low-level state. Rare exceptions then demand rapid reconstruction: the out-of-the-loop problem. Good supervision needs useful abstraction with drill-down, not every sensor stream.
Studying it
Teleoperation, supervision, and higher autonomy can compare performance, command frequency, workload, situation awareness, anomaly diagnosis, and takeover. Long quiet periods and changing goals matter, including goal-setting errors. Algorithm capability should remain comparable.
Where it stops holding
If goals and constraints cannot be expressed, automation optimises a poor proxy. Very short intervention windows or frequent unstructured change may need direct control. Nominal approval without information and time is not actual control.
Applying it
- Make goals, prohibitions, priorities, and completion criteria reviewable, with a plan summary before execution.
- Show milestones normally and escalate detail plus intervention when constraints or confidence degrade.
- Insert unannounced anomalies in long tasks, measuring discovery, diagnosis, takeover, and goal error.