Can a keepout cover open floor?

Yes. It can represent an operational restriction without a physical obstacle.

Does every keepout prohibit entry absolutely?

No. The source terminology includes both discouraged and forbidden regions, so the implementation must be checked.

Will a keepout automatically protect every part of a robot?

No. Footprint modeling and the navigation components' collision checks must account for its physical extent.

AI Glossary / Reinforcement learning and robotics / Intermediate

Keepout

Definition

A keepout is a designated area that autonomous robot navigation should avoid or must not enter. It is usually represented as a spatial annotation, mask, or navigation constraint. The precise effect depends on the system: some keepouts express a strong route preference, while others mark an area as forbidden to the navigation planner.

What it means

A robot's sensors may see an empty patch of floor that remains unsuitable for operation. It might be reserved for people, temporarily unavailable, or outside the approved working area. A keepout adds this operational meaning to the navigation representation.
The term does not necessarily describe a physical obstacle. It tells the navigation system how to treat a region. A soft discouragement and a hard prohibition have different consequences, so their meaning must be explicit when the zone is configured and tested.

How it works

An operator or application defines the zone in a known coordinate frame, often as a polygon or grid mask. Navigation software incorporates the annotation into the representation used to evaluate routes or robot motion.
A global planner needs the restriction to avoid proposing an unsuitable route. A local controller also needs compatible information so its immediate motion remains consistent with that route restriction. Nav2 provides a keepout costmap filter and documents using it in both global and local costmaps.
The robot occupies space beyond its reference point. Footprint checking, clearance handling, and filter configuration therefore matter at boundaries. A center point outside the drawn region can still leave part of the robot inside it.
The active map, localization estimate, annotation alignment, and updates must agree. A keepout implemented in ordinary navigation software is not, by itself, evidence of a certified protective function or a physical barrier.

Three practical examples

Facility operation: a delivery robot avoids a designated area used for temporary storage even when sensors detect enough open floor to pass.
Demonstration setup: a robot is constrained to a defined public demonstration space while other parts of a hall remain outside its navigation area.
Workflow management: a fleet avoids a work zone while an authorized operator changes equipment layout, then uses a reviewed map update when access is restored.

Think of it this way

Think of a shaded no-entry area on a route map. The road may physically exist, but the route planner should treat it according to the restriction. The shading must be correctly positioned and understood by the system using the map. Drawing it on a screen is only the beginning; the moving vehicle must actually respect it.

Worked example

Consider an original geometric example, not a deployment specification. A rectangular keepout occupies x from 2 to 4 meters and y from 1 to 3 meters. Model the robot as a disk of radius 0.3 meters and add an illustrative clearance margin of 0.2 meters.
The robot center should remain more than 0.5 meters from the rectangle to preserve that chosen margin. A horizontal centerline at y = 3.4 passes only 0.4 meters above its upper edge, so it fails the requirement. A line at y = 3.6 has 0.6 meters of separation above that edge and passes this simplified geometric check.
One conservative rectangular approximation expands the forbidden center region to x from 1.5 to 4.5 and y from 0.5 to 3.5. This approximation excludes extra space near the corners; the exact disk-expanded boundary has rounded corners. The approximation is therefore not identical to the exact geometric expansion.
For a start at (0, 2) and goal at (6, 2), the direct six-meter route crosses the keepout. An illustrative rectilinear detour via (0, 3.6) and (6, 3.6) has length 1.6 + 6 + 1.6 = 9.2 meters. It stays outside the simplified expanded rectangle but is not claimed to be the shortest feasible route.
At 0.1 meters per grid cell, the combined 0.5-meter radius and margin corresponds to five cells. Actual rasterization, turning motion, localization uncertainty, and braking behavior still require separate treatment. The example's numbers do not establish a universally appropriate clearance.

Illustrative case study

Fictional case: Juniper Logistics introduces a temporary keepout near an indoor packing station. The operations team records the zone's purpose and checks that its coordinates match the robot's active map.
Engineers inspect planned routes and conduct controlled tests of boundary approaches, rerouting, and goals placed inside the excluded area. They also check what happens when the temporary annotation is unavailable or a new map is loaded.
Once the work area changes again, an authorized operator reviews the annotation before updating it. This invented case illustrates configuration ownership and verification; it does not claim a certified safety system or a measured reduction in incidents.

Common misconceptions and mistakes

A keepout is not necessarily an obstacle detected by sensors. A high traversal cost may discourage entry without prohibiting it. Checking only the robot's center can miss footprint overlap, and simply adding a zone to one map component may leave another component unaware of it.
Do not assume that normal obstacle inflation automatically applies to every keepout filter; implementation details differ. A displayed zone can also be misplaced through coordinate or map errors. Finally, changing a keepout does not resolve an impossible task if the permitted region has no route to the goal.

Business use

Keepouts help facilities express where robots are permitted to operate and coordinate changes in shared space. Their value depends on clear ownership, reliable map handling, and predictable responses to blocked goals. Teams should distinguish operational preferences from mandatory restrictions and verify the selected implementation accordingly. A useful acceptance test checks actual routes and motion near boundaries, not just whether a colored region appears in an interface.

Check your understanding

In the worked example, why is y = 3.4 unsuitable even though it is above the keepout's upper edge?

Answer: The centerline is only 0.4 meters from that edge, less than the chosen 0.5-meter combined robot radius and clearance margin.

Related terms

Learn next

Learn how navigation systems represent coordinate frames and robot footprints. Then compare soft traversal costs with forbidden regions and inspect how the selected platform applies keepout information during planning and control.

Sources

PAL Robotics, PAL OS glossary, Keepout and Map Mask entries: https://docs.pal-robotics.com/edge/glossary Nav2, Keepout Filter Parameters: https://docs.nav2.org/rolling/configuration_and_development/configuration_guide/core_servers/costmap_2d/costmap_filters/keepout_filter/ Nav2, Navigating with Keepout Zones: https://docs.nav2.org/rolling/tutorials/general_tutorials/navigation2_with_keepout_filter/navigation2_with_keepout_filter/

Independent review checked PAL terminology and Nav2 filter configuration.

Previous: Karush-Kuhn-Tucker conditions Next: Kendall's coefficient of concordance ( W )