Who Is Allowed to Make the Decision?

Robotics is not only about whether a machine can make a decision.

It is also about whether the machine should own that decision.

Decision authority

Consider an autonomous mobile robot in a school hallway.

The robot may be allowed to:

• choose a local path;
• slow for an obstacle;
• retry a docking maneuver.

It should not automatically be allowed to:

• enter every room;
• move after an emergency stop;
• ignore a person blocking the path;
• decide that missing sensor data is safe.

Authority is part of architecture.

Human in, on, or outside the loop

These phrases are used in different ways across industries, but the underlying questions are useful:

• Does a human approve each important action?
• Does the robot act while a human supervises?
• Can a human interrupt?
• How quickly can the interruption take effect?
• What happens if the human interface fails?

Automation surprise

A dangerous interface is one where the robot changes mode and the person supervising it does not realize what changed.

Examples:

• teleoperation quietly becomes autonomous hold;
• manual override expires without clear indication;
• robot resumes after a fault clears;
• interface displays stale status.

The human model and machine state must agree.

Authority map

For one robot mission, create three columns:

Add at least six decisions.

Then identify one decision where your first instinct changed after considering consequence and recovery.

Authority should be explicit

Consider an autonomous mobile robot with three possible sources of intent:

• autonomous planner;
• remote operator;
• emergency stop.

If all three can issue motion commands without an authority rule, conflict becomes inevitable.

A simple priority model might be:

emergency stop
      ↓ overrides
manual safety control
      ↓ overrides
autonomous planner

The exact hierarchy depends on the application, but it must be deliberate.

Separate permission from capability

A robot may be technically capable of entering an area while not being permitted to do so.

A manipulator may be physically capable of exerting a force that violates the task's safety limit.

Architecture therefore needs both:

• capability constraints, based on what hardware/software can do;
• authority constraints, based on what the system is allowed to do.

Record the handoff

For one autonomous function, make an authority table:

A human-in-the-loop system is not safer merely because a person exists somewhere nearby. The interface must make intervention possible in time.