Disclosure — some links on this page go to Amazon, and we earn a commission if you buy through them. It costs you nothing extra and it does not change what we recommend or in what order. The full disclosure.
Obstacle avoidance runs on separate hardware from mapping — usually a forward-facing camera, a structured-light projector, or both. Roborock states its Reactive AI recognizes 108 object types. iRobot markets its avoidance specifically against pet waste. None is perfect, and the honest advice is still to pick up the cables.
This is the feature that determines whether a robot is a machine you set running or a machine you tidy for. It is also the one where the gap between marketing language and reliable behavior is widest.
What the hardware actually is
A forward-facing camera with object recognition. Sees what an obstacle is, not just that it is there — which is how a robot distinguishes a sock from a doorstop. Needs light, and is the reason cameras appear on robots that map with LiDAR.
Structured light. A projector casts a pattern ahead and a sensor reads how it deforms, giving a depth map. Works in the dark and is good at low, wide obstacles. Dreame states 3D structured light on the X50 Ultra; ECOVACS states a TruEdge 3D edge sensor alongside its dToF.
Both together. The current flagship approach: structured light for shape, camera for identity. Roborock's Reactive AI on the Curv line is this combination.
Published navigation and obstacle hardware by model
23 of 23 models publish this figure. Every number below links to the page it came from.
Compiled from the manufacturers’ own published material on 2026-09-01. We have not measured any of these figures ourselves — see how we pick. Found an error? Tell us and we will correct it.
What it reliably avoids, and what it does not
Reliable, on most current systems: shoes, cables lying loose, pet bowls, larger toys, furniture legs, the dock itself. Less reliable: thin cables against a matching floor, dark objects on dark floors, anything transparent, and low flat items like a folded sock that presents almost no profile.
The manufacturers' object counts — 108 types, and similar figures elsewhere — are a claim about the recognition model, not a guarantee about your floor. Treat them as an indicator of investment rather than a specification.
The trade-off nobody mentions
Aggressive avoidance leaves gaps. A robot that gives a wide berth to everything it detects will not clean the strip around a chair leg or the space beside a cable. Some models let you tune the behavior, and some let you disable avoidance entirely for a deeper clean of a floor you have already tidied. If your floors are usually clear, the tidier-and-closer setting genuinely cleans better.
How we picked
Criteria first, products second. These are the things we ranked on, stated before anything is recommended.
What avoidance hardware the maker actually publishes — a camera, structured light, or both.
Whether the maker states what it recognizes, rather than claiming general intelligence.
Whether the same hardware works in the dark, since most runs are scheduled when the house is empty.
How the model handles the highest-cost obstacle, which is pet waste.
Whether avoidance can be turned off, for people who would rather tidy than be routed around.
What we did not do: test any of these machines. We do not own one, we do not have a lab, and nothing on this page is written as though we did. The full method.
Quick picks
Ranked on the published specifications below. Tap a row to jump to the full write-up.
#
Product
Best for
Max climb
Price
1
RoborockNo photo
Roborock Qrevo CurvXThe highest published suction figure here paired with the highest published mop lift, which is the combination a mixed-floor home actually needs.
iRobot Roomba j9+The vacuum-only j9+ is the model whose obstacle avoidance is specifically marketed against pet waste, which is a failure mode with a very high cost.
ECOVACS DEEBOT X8 PRO OMNIA rotating roller mop that washes itself continuously while it cleans, instead of pushing one increasingly dirty pad around the house.
Dreame X50 UltraThe only model here that legs itself over a step rather than driving at it, and the only one that publishes the condition attached to its own headline number.
The tallest published climb figure
60mm over a series of steps, 42mm over a single step
The only model here that legs itself over a step rather than driving at it, and the only one that publishes the condition attached to its own headline number.
60mm over a series of steps, 42mm over a single stepDreame's own condition, quoted: "The gap between steps should be greater than 4cm. For a single obstacle crossing and a single vertical step, the highest obstacle crossing height is 4.2cm." The 6cm headline therefore does not describe a single door sill.Source: Dreame X50 Ultra product page — ProLeap step climbing (retrieved 2026-09-01)
Publishes the caveat as well as the headline, which is rarer than it should be
20,000 Pa stated peak suction and a 395ml on-board bin
Stated 220 minutes of runtime in quiet mode
What it does not
The legs add height and mechanical complexity to a machine that lives under furniture
10.5mm mop raise is modest next to the 17mm figures from Roborock
Who it is for
Split-level rooms, a sunken living room, or a doorway sill that has beaten every robot you have owned.
Skip it if
You read the 6cm headline and stopped there. Dreame's own footnote puts a single vertical step at 42mm, and 6cm only applies to a series of steps with more than 4cm between them.
It has no obstacle avoidance at all — it finds obstacles by driving into them, which is what a bump sensor is. That is a perfectly honest design at its price and it is the wrong machine for a floor with cables, toys or a puppy on it. Tidy the floor or buy a robot that looks where it is going.
Questions people actually ask
+ What is the best robot vacuum for obstacle avoidance?
One that publishes what its hardware actually is — a camera, structured light, or both — rather than only a marketing name. Roborock, Dreame, ECOVACS and iRobot all describe theirs; the comparison above shows what each states.
+ Does obstacle avoidance work in the dark?
Structured light does. Camera-based recognition does not work well without light, which matters because most people schedule runs when the house is empty and the blinds are shut.
+ Will a robot vacuum avoid dog mess?
The models that specifically market avoidance for it are much more likely to. None offers a guarantee, and no honest source will give you one. If you have a puppy or an older dog, this is worth buying for.
+ Should I turn obstacle avoidance off?
If your floors are genuinely clear, yes — avoidance leaves a margin around anything it detects, and turning it off cleans closer to furniture legs. If the floor has cables and toys on it, leave it on.
+ Why does my robot avoid something that is not there?
Usually a strong shadow, a high-contrast pattern in the flooring, or a reflective patch that reads as an object. It is a recognition false positive rather than a fault, and it often clears once the map has been rebuilt in better light.
