Start with a tape measure, not a spec sheet

Published threshold-crossing heights across these eight models run from 0.94 in to 3.47 in, and two manufacturers publish no figure at all. That is a wider spread than any other factor here, and it is the only one you can settle before ordering by walking around your own house.

Read the single-layer number, not the headline

Where a manufacturer publishes both, the two figures differ by a lot. Roborock rates the Qrevo Edge at 4 cm for double-layer thresholds and 3 cm for a standard single one. Ecovacs rates the X11 at 1.57 in for multi-layer sills and sliding-door tracks but 0.94 in for a single threshold. Roborock’s 3.46 in for the Saros 20 is explicitly a stacked pair, 1.77 in on top of 1.69 in.

A double-layer rating describes two obstacles in sequence, such as a transition track with a rug edge sitting on it. If the spot that worries you is one raised sill under a bedroom door, the single-layer figure is the one that applies, and it is the smaller of the two. Buyers who compare headline numbers across brands end up comparing a stacked figure from one against a single-threshold figure from another.

Lift, retract, shield: same problem, three mechanisms

Every model here states that it reacts to carpet automatically, but what it does and whether it publishes a number vary.

Three publish a height. Narwal states the Flow 2’s rolling track mop lifts 12 mm on carpet detection, the highest lift figure in this comparison. Roborock states the Qrevo Edge raises the whole chassis 10 mm and detaches the mop. Ecovacs states the X11’s Triple Lift System raises the roller mop 10 mm, triggered by ultrasonic carpet detection.

Five describe the behavior without a figure. Roborock’s AdaptiLift on the Saros 20 auto-detects carpet and lifts the mop; Dreame states the X60 Ultra’s pads lift automatically; iRobot’s retractable cover shields the pad; Shark’s Floor Detect and NeverStuck lift the pad at floor-type changes; MOVA’s AutoShield retracts the mop and deploys shields. These are plausible mechanisms and several are more elaborate than a simple lift. They are also unverifiable before purchase, which is why they score lower here than the published 10 mm and 12 mm figures.

One more distinction is worth noticing. Ecovacs is the only manufacturer among these eight that names the sensing method behind the trigger, ultrasonic detection. Everyone else says “on carpet detection” or gives the feature a name.

Peak suction is the number people optimize and it decides the least

The published peaks: 36,000 Pa for the Saros 20, 35,000 Pa for the Dreame, 31,000 Pa for the Narwal, 30,000 Pa for the iRobot but only on the Amazon listing rather than iRobot’s own documentation, 28,000 Pa for the MOVA, 19,500 Pa for the Ecovacs, 18,500 Pa for the Qrevo Edge, and nothing at all for the Shark.

What none of them publish is a carpet-mode suction figure in Pa. Two get closer than the rest: Narwal rates its pressurized CarpetFocus plate at a 4.2x instant suction increase on carpet, and MOVA states the Z60 holds its full 28,000 Pa and keeps the side brush turning while crossing carpet rather than pausing the clean. Roborock is the only one to publish a carpet pile-height limit for the Saros 20’s Dynamic Cleaning, which adjusts chassis, brush and mop height together.

So a peak figure tells you about the motor, not about what happens when the robot rolls off oak onto a rug. The behavior descriptions do, and they are what to compare.

The exception worth flagging is the one product that makes the response manual. Roborock states the Qrevo Edge’s mid-to-long-pile carpet mode is turned on in the app. If you bought a robot specifically so nobody has to think about the floor changing, that is a design difference, not a detail.

Brush design is where hair actually stops a cycle

One measurement exists in this set. Independent testing published by Vacuum Wars measured 0% hair wrap on the Saros 20’s DuoDivide brush, which uses a center gap to channel hair, against a 26% category average. That is checkable in a way no manufacturer claim is.

Below that, the descriptions vary in how much they commit to. Ecovacs specifies a floating ZeroTangle 3.0 brush with a Triple-V structure and spiral-comb integration. MOVA specifies a pure rubber brush paired with a rubber-bristle brush and rates it at 99% hair removal, its own figure rather than an independent one. Roborock’s Qrevo Edge uses dual parallel short-bristle rollers. iRobot lists dual rubber brushes with dual extending edge-sweeping brushes. Dreame, Narwal and Shark name their brushes, HyperStream Detangling DuoBrush 2.0, Zero-Tangling Floating Roller Brush and Self-Cleaning Brushroll, without describing the construction. A named brush is not the same evidence as a described one.

Narwal’s Flow 2 is described with dual 1080P cameras and a VLM processing brain, and no LiDAR unit is mentioned on the product page. MOVA describes FlexScope with edge-laser wall and corner following but does not state whether the underlying system is LiDAR or camera based, the only model here that leaves that unnamed.

Two designs solve a real conflict between mapping and furniture. Dreame’s VersaLift retracts the LiDAR flush into the body, and Roborock’s StarSight 2.0 puts the LiDAR inside the chassis with no top turret, letting the Saros 20 navigate as low as 3.13 in. In a home where the robot has to get under a bed to reach the carpet beneath it, that matters more than the brand of sensor.

When the shorter-climbing option is genuinely enough

If every transition in your home is flat or a low strip, the 1.57 in group is not a compromise. The Narwal, the Qrevo Edge and the Ecovacs all publish figures at or near that mark, and the decision moves entirely to carpet behavior, brush design and dock maintenance, where Narwal’s four named carpet modes and Ecovacs’ ultrasonic trigger are the strongest published records here.

It stops being enough the moment one doorway exceeds the published figure. A robot that cannot get over the sill does not clean the room behind it, and no amount of suction changes that. That is also the argument against the two models with no published height: with the iRobot and the Shark you cannot run this check at all, which is a documentation problem rather than a verdict on the hardware.

The dock spec that matters depends on your floor ratio

Docks in this set are described mostly by wash temperature, which is a mopping spec.

Runtime is published by only two. Ecovacs states 207 minutes sweeping and 206 minutes sweeping and mopping; Shark states up to 2 hours in dry vacuum mode on hard floors with a 4-hour charge. For a large open-plan floor with several surface changes, that is a more relevant published figure than peak suction, and six of these eight do not give it to you.