Can an open-loop command change over time?

Yes. A predetermined sequence can vary over time without using output feedback to revise it.

Does logging the output make a system closed-loop?

No. The logged information must influence the relevant control action for that feedback path to be closed.

Can one system contain both types?

Yes. Different layers or subsystems can use different feedback arrangements, so the label should specify its scope.

AI Glossary / Graphs and symbolic AI / Beginner

Open-loop

Definition

Open-loop describes a control or action process that does not use feedback from the controlled result to revise its commands during the relevant execution period. The system acts according to a schedule, model, or predetermined sequence. In AI planning, an open-loop plan follows its selected actions without adapting them to new observations of their effects. The scope of the loop must be stated clearly.

What it means

A timed watering device can run for ten minutes without measuring how much water reached the plants. Its action is based on a planned duration, not on the achieved result. If the flow changes, the timer does not automatically compensate.

Open-loop does not mean that no information was used. The schedule may have been calculated carefully from measurements taken earlier. Nor does it mean that the command is constant: a changing sequence can still be open-loop if its changes do not depend on feedback about the controlled outcome.

How it works

The controller selects inputs using a desired target, an assumed process model, and any information allowed by its design. It then sends those inputs to the process. The resulting output does not return through a feedback path that corrects the relevant command.

Performance therefore depends on how accurately the assumptions describe the process and its disturbances. A predictable process may follow the planned trajectory closely. An unexpected load, external disturbance, or incorrect starting condition can make actual results differ from the intended ones.

The system boundary matters. A high-level robot plan can be open-loop while a low-level motor controller uses feedback to regulate speed. Calling the entire machine open-loop would conceal this distinction. Similarly, a monitoring sensor that only records results does not close a control loop unless its information influences the relevant action.

Three practical examples

Timed dispensing: run a pump for a preset duration without adjusting to measured delivered volume.
Scripted simulation: execute a fixed action sequence without changing it when the simulated state differs from the prediction.
Automated process scheduling: issue a predetermined sequence of commands while separately logging outcomes for later review.

Think of it this way

Think of cooking by a fixed timer without checking the food. The instructions may work under familiar conditions, but a different starting temperature or appliance behavior can change the result.

The analogy is limited because real control designs can combine planned actions with feedback at several levels. The relevant question is exactly which measured outcomes can change which commands, not whether the system includes any timer or sensor at all.

Worked example

Suppose a toy tank starts with 2 liters and should reach 8 liters. A pump is assumed to deliver 0.5 liters per minute. An open-loop timer therefore runs it for (8−2)/0.5=12 minutes.

If the actual flow is 0.4 liters per minute, the delivered amount is 12×0.4=4.8 liters. The tank ends with 6.8 liters, which is 1.2 liters below the target. The timer completes successfully even though the intended volume is not reached.

If instead the actual flow is 0.6 liters per minute, the same command adds 7.2 liters and the final amount is 9.2 liters, 1.2 liters above the target. The controller has no output-based correction in either case.

A feedback design could measure volume and alter the pumping decision, but its accuracy would depend on measurement quality, delay, and how the pump responds. This arithmetic assumes constant flow, no leaks, and no capacity limit. It is an original teaching example, not a specification for a physical filling system or proof that any particular feedback controller will meet the target.

Illustrative case study

Fictional case: a small sorting robot follows a fixed sequence of motions that works in a carefully arranged demonstration. When a box begins a few centimeters away from its expected location, the same sequence misses its intended pickup point.

The team documents which parts of the sequence depend on exact initial conditions and considers where observation should influence action. It distinguishes position feedback within the robot's motors from feedback about the box's location. The case illustrates different control levels, not a claim that all scripted automation is unsuitable.

Common misconceptions and mistakes

Open-loop is not synonymous with inaccurate. A well-calibrated process in stable conditions can perform adequately without output feedback. Suitability depends on the tolerances and disturbances of the task.

It is also not synonymous with having no sensors. Sensors may be used for logging, initial calibration, or another subsystem while the particular control path remains open-loop.

Closed-loop control is not automatically perfect either. Feedback can introduce its own problems, including instability when poorly designed or sensitivity to noisy measurements. The meaningful comparison concerns a specified task, model, and operating conditions, rather than a blanket ranking of the two labels.

Business use

For business automation, identifying open-loop behavior helps expose where the system assumes its actions have the intended effects. This is useful when evaluating scripts, robot demonstrations, and processes that run for a fixed time.

Teams should state acceptable error, expected variation, and what evidence establishes completion of the real task. A command reporting that it finished is not always evidence that the desired physical or business result occurred. Where feedback is added, its role, reliability, and influence on action should be explicit and tested within the complete process.

Check your understanding

Why does the toy pump stop at twelve minutes even when the tank holds only 6.8 liters?

Answer: Its stopping command depends on the preset timer, not on a measurement of the achieved tank volume.

Related terms

Feedback
Execution monitoring
Closed-loop control
feedforward control
open-loop plan
control system
disturbance
state estimation

Learn next

Draw the flow from command to process to result in a simple automation. Mark which measurements return to change the command. Then distinguish feedback at the task level from feedback inside individual actuators or software components.

Feedback

Sources

Åström and Murray, Feedback Systems, Introduction: https://fbswiki.org/wiki/index.php/Introduction — open and closed interconnections, feedback benefits, and design limitations.

Russell and Norvig, author-hosted search chapter: https://aima.cs.berkeley.edu/4th-ed/pdfs/newchap04.pdf — executing a completed plan without recourse to new percepts. The tank calculation and sorting scenario are original illustrations.

Previous: Open-label trial Next: Open set recognition