A perfect rectangle around the pet bowls can still let your vacuum nose 10 inches onto the rug. That doesn’t mean the robot ignored you — a no-go zone is a software line on a map the robot keeps guessing at, not a fence it can feel.
The guess holds steady on open, light flooring, but it wobbles where the floor changes color, reflects, or goes dark. On those edges the robot thinks it is still outside the zone while you can see it inside. That gap lines up with what manufacturers themselves describe: iRobot Keep Out Zone difficulty says entering a Keep Out Zone often means the robot is having difficulty understanding where those zones sit inside its map, and recommends starting from the dock with good lighting and a clear camera. Once you measure the real overshoot with tape on three surfaces, you can tell in seconds whether you need more light, a different boundary spot, or a different navigation type — and that check is what the rest of this guide walks through.
Why a drawn line in the app is not a wall on the floor
When you draw a no-go zone, the app saves a polygon on a saved floor plan. While cleaning, the robot constantly estimates where it is inside that plan by matching what its sensors see right now to that saved map. That estimate is never exact — it lives inside an uncertainty ellipse that grows and shrinks.
LiDAR models use an active laser to measure distance to walls. Camera models use visual features and probabilistic localization to refine position through successive guesses. In both cases, the no-go check is simple: is my estimated position inside the polygon? If the estimate drifts, the check passes even though the bumper is physically over your line.
That drift gets worse at specific spots. Dark rugs absorb infrared light, so cliff sensors can read them as a drop and visual features disappear. Mirrors and glass doors create ghost points because the laser sees a reflection beyond the glass. Low light starves a camera of texture. Dreame positioning cause lists this as a positioning puzzle — the robot is still figuring out its place before it enforces a boundary. Narwal guide notes the same: uneven lighting or reflective surfaces cause inconsistent performance across rooms, along with outdated maps and dirty sensors.
Diagram showing LiDAR laser and camera inputs feeding probabilistic map, confidence drop near dark and reflective edges, and where no-go check occurs against estimated position
If you see the app trace staying just outside the rectangle while the real robot sits on the rug, that’s localization uncertainty, not disobedience. The fix starts with where confidence drops, not with redrawing the line tighter.
LiDAR vs camera accuracy and why that changes your boundary drift
LiDAR and camera systems both build a probabilistic map, but they lose confidence for different reasons. LiDAR shoots its own light, so it works in the dark and typically holds millimeter-level map precision on normal walls. Camera-based navigation, often called vSLAM, needs adequate light and visual texture, and its accuracy shifts with lighting and surface contrast.
That shows up in manufacturer specs and independent tables. Roborock’s StarSight system, for example, lists Roborock sensor points at 21,600 points for 3D scanning, which defines the resolution tolerance its map is built from. A head-to-head comparison table summarizes it plainly: LiDAR millimeter precision is high via laser scanning, while camera is moderate and depends on lighting and textures.
In practice that means LiDAR will usually keep a boundary on a dark rug edge better than a camera in a dim room, because color doesn’t absorb its laser the way it does visible light. But LiDAR can still be fooled by a full-height mirror or glass door that creates a second room in the laser data. Camera models struggle more on a uniform dark rug where cliff sensors read dark as a drop and the camera sees little texture to lock onto.
On the official Roborock owner forum, support staff make the same distinction this article does: an invisible wall only blocks one approach angle, while a no-go zone blocks all of them — so a boundary that “fails” from one direction may simply have been the wrong tool for that spot, not a drift problem at all. The forum’s repeated advice for a boundary that keeps getting nicked is the same fix that shows up across owner communities generally: watch the app’s live trace during a run and move the boundary 6 inches outward from the actual hazard, because the mechanism is uncertainty, not the app ignoring the line.
Interactive table comparing LiDAR and camera navigation accuracy across light, dark, and reflective conditions and resulting no-go zone drift risk
How to measure boundary drift with a tape measure on three surfaces
Marketing claims like precision mapping sound absolute. A tape measure makes it concrete. Here’s the method: draw an identical 24×24 inch no-go rectangle in the same map location, mark its edge on the floor with painter’s tape measured from a wall, start the robot from its dock, and measure the furthest point of the bumper past the tape where it stops or turns.
Repeat three runs per surface so one lucky run doesn’t fool you. You’ll need painter’s tape, a tape measure, and three spots that exist in many homes: a light hard floor, a dark rug edge where the rug meets that floor, and a reflective edge near a mirror or glass door. Clean the LiDAR window, camera lens, and cliff sensors first, start from the dock each time with adequate lighting for camera models, and save the map before testing.
Digital Trends boundary setup methods shows both physical tape and app virtual walls as valid boundary methods — useful for anchoring this test because the tape gives you a physical line that matches the app line.
Checklist showing tape-measure boundary drift test — mark line, start from dock, measure overshoot on three surfaces, log results
At your no-go edge, look for: does the bumper stop on the tape on light floor but roll 2 inches past it on the dark rug? That pattern points to mapping confidence, not an app bug.
Table comparing measured boundary overshoot across light, dark, and reflective surfaces with interpretation for no-go placement
This rubric is a practical evaluation tool created for this guide based on sensor confidence by surface type, lighting for camera models, and reflective ghost mapping described above, not a published industry standard. Use it as a quick in-home check.
What your drift numbers tell you about LiDAR vs camera limits
If light floor holds under 1 inch, that’s a healthy result — it’s in line with the centimeter-level wall and furniture-leg precision that current 3D ToF and LiDAR navigation systems are built to deliver. It means your map is healthy.
If dark rug edge shows 2 to 4 inches while light floor stays tight, you’re seeing cliff sensor absorption and low visual feature density. LiDAR is less affected by color, but camera models will drift more here unless lights are bright and the floor is decluttered. If reflective glass pushes drift past 6 inches, that’s not color — it’s the laser or camera seeing a second room in the mirror and stretching the map.
Cross-check: if observed drift exceeds your manual’s stated tolerance at a specific spot but not elsewhere, confidence degraded at that spot, not everywhere. That points to placement or lighting, not a defective unit.
Why no-go zones fail even when the map looks right
A map can look perfect and still let the robot cross. Beyond surface confidence, four other failures show up often.
First, map freshness. Move a couch or pet bowls and the saved map no longer matches what sensors see, so localization drifts. Second, zone logic: overlapping no-go rectangles, or a no-mop zone mistaken for a full exclusion, can create gaps. Third, Wi-Fi save: if Wi-Fi drops while you tap save, the zone never commits. Narwal common issues lists outdated mapping data, unclear boundary lines, and dirty sensors together for this reason.
Fourth, the robot was lifted or moved. Xiaomi localization failure notes that if the robot encounters an error or fails to locate itself, it may ignore restricted areas, and warns not to lift or relocate during cleaning. Vorwerk documentation notes a similar threshold: moving more than about 0.5 meters loses orientation. When that happens the robot is literally lost, so it can’t enforce any software line.
The last one is dirty optics. A dusty LiDAR window or smudged camera lens lowers feature quality everywhere, so even light floors start drifting.
The mistake that makes every fix look like it didn’t work
The common fix that backfires is drawing the no-go tighter to the hazard after it drifts. If the robot’s uncertainty ellipse is 3 inches wide at that rug edge, pulling the line inward by 2 inches just means it now crosses even deeper. It looks like the robot is ignoring you more, when actually the same uncertainty is still pushing it over.
A better move is counterintuitive because the app line looks precise: draw the boundary 6 to 12 inches outward from what you actually want to protect, and keep at least 8 inches clearance from mirrors or glass doors. You give the confidence wobble room to live in. The app still shows a crisp line, but real edge accuracy depends on confidence at that spot, not on line art.
How to make a virtual boundary actually hold
Start from the dock every time. iRobot fix steps specifically call out starting from Home Base, adequate lighting, and a camera not obstructed, and recreating the Imprint Smart Map if the robot can’t navigate to a room. That start point gives the robot its best known position before it enforces any zone.
Then clean the optics. Wipe the LiDAR turret window, camera lens, and cliff sensors with a dry cloth. Delete overlapping zones — keep one simple rectangle per hazard — and tap the checkmark or save button while on strong Wi-Fi. Confirm the correct floor level is selected if you have multi-floor maps.
Place boundaries where confidence is high, not where the hazard is. For a dark rug edge, place the line on the light hard floor just before the transition, not on the rug itself. For reflective, cover a full mirror during mapping or map with curtains closed, then add a physical magnetic strip as backup if your model supports it. Narwal maintenance tips suggest keeping maps updated after furniture moves and testing regularly, which catches drift before you need it.
If your home is dim by design and you run the vacuum at night, a LiDAR model will typically hold a no-go tighter than a camera model with lights off, because LiDAR doesn’t need ambient light. If you keep a camera model, run it with lights on and declutter the floor so it has texture to lock onto.
Try this before you buy a different model: recreate the map in bright light starting from the dock and retest drift with tape — if drift shrinks, it was localization, not hardware.
The check that changes the decision
A no-go zone is a software line drawn on a probabilistic map, so it only holds where that map’s confidence is high. The tape test shows this in minutes: same rectangle, three surfaces, start from dock each time.
The most important action is running that same-boundary test on light hard floor, dark rug edge, and reflective glass and logging overshoot. If you skip the surface-specific test, you’ll keep redrawing the line tighter and assume the robot is defective, when moving the boundary outward or adding light actually fixes it.
Frequently Asked Questions
Why does my robot vacuum cross the no-go zone only on my dark rug?
Dark surfaces absorb light that cliff sensors use, so the robot can treat the rug as a drop and its camera sees fewer features, which grows pose uncertainty at that edge. Draw the boundary on the light floor before the rug transition and clean cliff sensors. Narwal notes inconsistent performance from uneven lighting or reflective surfaces, while iRobot guidance calls for adequate lighting and clear camera.
Does LiDAR or camera navigation have better no-go zone accuracy?
LiDAR typically holds LiDAR vs camera accuracy at high millimeter precision independent of light, while camera is moderate and depends on lighting. Both can drift near mirrors. Look for stated resolution like sensor points at 21,600 as the tolerance basis, and prefer LiDAR for dark homes.
How do I fix a no-go zone that keeps moving or not saving?
Tap the checkmark to save, stay on strong Wi-Fi, start from dock, delete overlapping zones, and select the correct floor map. If the map looks distorted, remake it. Dreame causes include map mix-up and positioning puzzles, while Xiaomi notes localization fails if lifted mid-clean.
