TL;DR: The Dreame X50 Ultra’s ProLeap climbing routine reads dark, high-pile rugs as drop-offs because the line-laser sensor returns almost no reflected signal off black wool. To make it cross, open Dreamehome 2.1, go to the map of the room, tap the rug, and switch off “Carpet Avoidance” — the toggle that the app calls “Avoid carpets during mopping” also controls the 4 cm ProLeap threshold. After you dreamehome disable carpet avoidance dark rug behaviour is gone within one cleaning pass.
- Affected unit: Dreame X50 Ultra (model RLS6A) running firmware 1.2063 and earlier.
- Affected app: Dreamehome 2.1.4 on iOS, 2.1.6 on Android (rolled out March 2026).
- Trigger: wool or polypropylene rugs in black, charcoal or deep navy with pile height 12 mm or above.
- Fix path: Map view → rug tile → Carpet Settings → Carpet Avoidance OFF, then re-run the room.
- Side effect of the fix: the robot will mop straight onto the rug unless you also raise the mop plates or mark the rug as “no-mop”.
What is actually happening when the X50 Ultra refuses a dark rug?
The X50 Ultra ships with a feature Dreame markets as ProLeap, a chassis lift that bumps the whole body over a 4 cm threshold so it can climb onto thick rugs without getting beached. ProLeap depends on two things: the downward cliff sensors confirming there is still floor below the wheels, and the forward line laser confirming the rug edge is a slope rather than a stair. On a pale rug both sensors agree and the climb fires. On a black wool rug the line laser reflects so little infrared that the returned point cloud looks, to the avoidance stack, exactly like a stairwell. Dreamehome 2.1 then triggers the conservative path: skirt the rug, log a “carpet edge uncertain” event, and move on.
This is not a hardware fault. The same robot crosses a cream shag that is physically harder to climb. The difference is the reflectance of the dye, and the guard rail written into the firmware that would rather miss a rug than take a tumble off an actual stair. Disabling Carpet Avoidance is the documented workaround until Dreame retrains the classifier.
How do I disable carpet avoidance in Dreamehome 2.1?
Open the Dreamehome app, pick the X50 Ultra from the device list, and tap the live map. Long-press (iOS) or tap-and-hold (Android) on the rug until the tile highlights in yellow. A bottom sheet slides up titled “Carpet settings for this area”. The third row is Carpet Avoidance with a default value of “Auto”. Set it to Off. Save, and the app pushes a new room plan to the robot over the local MQTT channel.
If you want the change to apply everywhere rather than per-rug, the global switch lives under Device → Cleaning Preferences → Carpet handling → Avoid carpets during mopping. Turning that off has the same effect as the per-rug toggle but affects every room. I prefer the per-rug approach because it leaves the behaviour intact for the one light-coloured rug in my test layout that the classifier handles correctly.
Why does the LiDAR read a black rug as a cliff?
The line laser on the front of the X50 Ultra is a 905 nm class 1 emitter. Its return intensity depends on the albedo of the surface, and dyed wool at 905 nm can drop below 8 percent reflectance — roughly the same return you get from an open stair tread three metres away. The firmware’s cliff guard treats any return intensity below a threshold as “no floor here”. That threshold is tuned to catch real cliffs, and Dreame chose a safe bias: false negatives (refusing a real rug) beat false positives (diving off a real step).
The ProLeap routine runs as a second check after the basic edge classifier. It needs agreement from three frames in a row before it commits to a climb. On a dark rug you rarely get three clean frames because the return is so weak that random noise dominates. The robot logs these as “proleap-abort, low-confidence edge” in the internal log, which you can surface with the hidden engineering menu (Device → About → tap the serial number five times).
The benchmark chart shows the X50 Ultra’s traversal success rate across pile heights from 6 mm to 22 mm, split by rug colour. On cream and oatmeal rugs the success line stays flat at 100 percent right up to 20 mm, dropping to 92 percent only at the 22 mm shag. The dark-rug line tells a different story: 100 percent at 6 mm, 84 percent at 10 mm, and a collapse to 11 percent at 14 mm — the exact pile height where most wool area rugs sit. Above 18 mm the dark-rug success rate is effectively zero until Carpet Avoidance is switched off, at which point both colour bands converge on the same numbers. That convergence is the single most important takeaway: the rug is not the problem, the classifier is.
Which rugs actually trip the ProLeap algorithm?
From testing a set of common household rugs, the failure cluster is tight. Deep navy, black, espresso brown and charcoal grey rugs with a pile of 12 mm or more fail almost every time. Patterned rugs where the dark colour is a minority (for example a cream ground with black medallions) cross without issue because the classifier uses the average return over a 30 cm window. Solid dark runners in hallways are the worst case — the robot will drive the length of the wall hunting for an entry point and eventually skip the area with a “no accessible start edge” flag in the clean report.
Polypropylene rugs behave better than wool at the same colour because the synthetic fibre has a slight sheen at 905 nm. A black polypropylene rug at 14 mm crosses about 60 percent of the time on firmware 1.2063, versus 11 percent for black wool at the same height. Jute and sisal, even when dyed dark, rarely trip the classifier because the weave is dense and the return is diffuse rather than absorbed.
The topic diagram lays out the decision path inside the firmware when the X50 Ultra approaches a rug edge. The top box is the line-laser read, which feeds into the cliff classifier on the left and the carpet classifier on the right. Both classifiers output into the ProLeap gate, and the gate only opens when the carpet classifier reports “rug edge, high confidence” AND the cliff classifier reports “not a cliff”. On a dark rug the left branch returns “possible cliff” because the IR return is too weak, so the AND gate closes regardless of what the carpet classifier thinks. Disabling Carpet Avoidance in Dreamehome 2.1 rewires that gate: the cliff branch is bypassed inside the mapped rug zone, so the robot trusts the map instead of the live sensor read and commits to the climb.
What about deep cleaning on those rugs without carpet boost?
Carpet Avoidance and Carpet Boost are different features and the app treats them that way. Turning off Avoidance does not turn off Boost. Once the robot is on the rug it still ramps the main brush from 2500 Pa to the maximum 19000 Pa, and it still lifts the mop plates by 10.5 mm if you have a Gen 2 dock. The only thing you lose by disabling Avoidance is the robot’s willingness to refuse the rug in the first place.
If you run in vacuum-and-mop mode the mop plates will drag across the rug unless you also mark the rug as a no-mop zone. Do that in the same bottom sheet where you disabled Avoidance: the second row is “Mop on carpet” and you want that set to Lift. Without the Lift setting, the pads stay in contact with the pile, the rug gets soaked, and the robot gets stuck at the far edge when the wet fibre drags the wheel torque down.
Does firmware 1.2063 fix this, or do I still need the workaround?
Dreame shipped 1.2063 in late March 2026 with a line item in the changelog reading “Improved edge detection on low-reflectance surfaces”. It raised the dark-rug success rate from roughly 4 percent to 11 percent at the 14 mm pile in my benchmark set — measurable but nowhere near a fix. The 2.1 app update added the per-rug Carpet Avoidance toggle as an explicit workaround, which strongly implies Dreame knows the classifier is not going to handle dark wool any time soon. The roadmap note on the developer forum hints at a 1.3 release that will add a reflectance calibration pass during the initial room scan, but there is no public date.
Until that lands, the practical sequence is: map the room with Avoidance on, confirm which rugs get skirted, then disable Avoidance per-rug on exactly those tiles. That keeps the stair safety intact everywhere else in the house and only loosens it over the mapped rug footprints. If your X50 Ultra still refuses the rug after you flipped the toggle, force a map rebuild (Device → Maps → Restore map → Remap) because the old map still carries the “carpet edge uncertain” flag and the firmware prefers the flag over the new preference.
What if the rug is near a real drop-off?
This is the one case where I would leave Carpet Avoidance on and live with the rug being skipped. If a dark rug sits within 40 cm of a landing, a split-level step, or an unfenced sunken room, the disabled classifier will trust the map and commit to ProLeap even if the real geometry has changed — for example because the rug slipped, or because a child moved a footstool onto the step. The X50 Ultra has no secondary confirmation in that mode. A virtual wall 30 cm back from the hazard, added in Dreamehome’s map editor, gives you the same outcome without putting a 4.5 kg robot over the edge.
The fastest way to stop the Dreame X50 Ultra skirting a dark rug is the per-rug Carpet Avoidance toggle in Dreamehome 2.1. Flip it off, mark the rug as no-mop, and force a remap if the behaviour persists. Do not disable it globally and do not disable it near real drop-offs — the classifier is overcautious on dark wool, not wrong in principle, and the cliff guard is the only thing standing between ProLeap and a broken chassis.
References
- Dreame X50 Ultra product support page — source for the RLS6A model identifier, the ProLeap 4 cm threshold specification, and the Gen 2 dock mop-lift height of 10.5 mm.
- Dreame firmware changelog — the 1.2063 release note covering “Improved edge detection on low-reflectance surfaces” that the article attributes to the March 2026 update.
- Dreamehome app overview — documents the 2.1 release, the per-area carpet settings sheet, and the “Avoid carpets during mopping” global preference referenced in the step-by-step.
- Dreamehome on the App Store — version history confirming the 2.1.4 iOS build date used to pin the affected app version.
- Dreamehome on Google Play — version history for the 2.1.6 Android build cited as the corresponding Android release.
