Robot Vacuum Autonomy Tier List: how we score hands-off cleaning

The Dust Rocket Autonomy Score is a single number, from 0 to 10, for how hands-off a robot vacuum is. The higher the score, the less you have to do: empty it less often, rinse mop pads less often, rescue it from under the couch less often. It is built entirely from the specifications we already track, so you can check our working on any product page.

This is not a review and it is not an opinion. We do not test vacuums in a lab, and the score has nothing to do with how well a machine cleans a given floor - it is a specification-based score, not a reliability verdict. It is our own data maths over published specs: every point is earned by a listed feature, and the product page shows exactly which parts counted and how many points each one earned. Two people reading the same spec sheet should land on the same score.

How the score is built

We add up six parts. Four of them can earn up to 2 points each and two can earn up to 1, and together they sum to a genuine 10. Each part scores the degree of a feature, not just whether it is present. Almost every robot vacuum sold today has a mop and a self-empty dock of some kind, so a simple feature count would bunch everything at the top and tell you nothing. Scoring the degree spreads machines out and makes the number useful.

1. Navigation and avoidance - up to 2 points

This is the single biggest factor in whether a vacuum can be left to its own devices, so it carries two points, split across two different jobs.

The first point is mapping: laser mapping (LiDAR, LDS or dToF) earns a full point, camera-based vSLAM or any other published mapping system earns half a point, and a machine with no published navigation earns nothing. The second point is obstacle avoidance: a machine that sees and steers around a stray sock or a cable using an AI camera or vision system earns a full point, structured-light or laser avoidance (which reads shapes but not what they are) earns half a point, and no avoidance earns nothing.

We score mapping and avoidance separately on purpose. Mapping is what lets the vacuum drive in tidy rows and clean room by room; avoidance is the cables-and-socks problem that decides whether you have to tidy the floor before every run. Mapping also sets a hard ceiling on the tier, which we explain below. Note that we do not reward the type of laser - we grade what the specs do for you, not the marketing generation of the sensor.

2. Dock - up to 2 points

A dock that empties the robot's bin for you earns 1 point. A dock that also washes the mop pads earns 2. A plain charging dock with no self-empty earns 0.

There is a higher tier of dock on the market - ones that wash with hot water, heat-dry the pads and refill the robot's clean-water tank automatically. We are not scoring those extras yet because we do not capture them as structured specs across enough models to be fair about it, so the dock part is currently capped at 2. When we do add them, it ships as a re-weight that keeps the scale at 10, not a change to the denominator.

3. Mop - up to 2 points

Active mopping that spins, rotates, rolls or scrubs earns 2 points - these actually apply pressure and motion to the floor. A mop that only vibrates or oscillates earns 1. A flat pad dragged along behind the robot, or no mop at all, earns 0.

4. Thresholds and climb - up to 1 point

This part is about getting over rug edges and door thresholds without getting stuck. A machine that can climb a 40mm step or higher earns a full point, 25mm to 39mm earns half a point, and anything lower earns nothing.

5. Power and endurance - up to 2 points

One point for suction and one for battery. On suction, 24,000 Pa or more earns a full point, 12,000 to 23,999 Pa earns half a point, and below 12,000 Pa earns nothing. On battery, a run time of 240 minutes or more earns a full point, 150 to 239 minutes earns half a point, and below 150 minutes earns nothing. A machine that is both strong and long-legged collects the full 2.

6. Upkeep savers - up to 1 point

A HEPA filter earns half a point and an anti-tangle brush earns half a point, for 1 point if a machine has both. Small things, but they are the difference between emptying a canister into a sealed bag and picking hair off a roller by hand.

Mapping is also a gate

Mapping earns its point in the score, but it does a second job as well: it sets a ceiling. A vacuum with real mapping - LiDAR or camera-based vSLAM - can reach any tier its score earns. A vacuum with no published navigation system bumps around semi-randomly, so no matter how strong its other specs are, we cap it at Tier 1. A machine you cannot trust to map a room is not really hands-off. The point rewards mapping in the score; the gate stops a machine with strong specs but no mapping from reading as hands-off.

When a high score still needs a hand

A single number can hide a deal-breaker, so we show the spec-derived ones in plain language next to the score rather than letting the figure over-promise. The main one is obstacle avoidance: a machine with none earns a hands-off limitation note telling you that you may need to clear cables, socks and toys before it runs. It is drawn straight from the specs, so it never claims a reliability verdict we have not tested.

From score to tier

The 0-to-10 score sorts every machine into one of five tiers, so you can shop by how much babysitting a vacuum needs rather than squinting at decimals:

• Tier 4 - Very high automation - a score of 9 or higher.
• Tier 3 - High automation - 7.0 to just under 9.
• Tier 2 - Moderate automation - 5.0 to just under 7.0.
• Tier 1 - Basic automation - 3.0 to just under 5.0, and the ceiling for any machine with no published mapping.
• Tier 0 - Minimal automation - below 3.0.

The headline is always out of 10, and it stays that way. The six parts add up to a real 10, so there is no artificial ceiling below the top. The denominator is permanent; the way the parts are weighted and where the tier lines sit are versioned, and any change is written up in the changelog at the bottom of this page.

Two worked examples

These show how the parts add up. They are illustrations of the method, not picks - for real machines and their live scores, see the ranked list below.

A flagship with the lot. LiDAR mapping and an AI-camera that dodges obstacles (navigation and avoidance 2), a self-empty dock that also washes the mop (dock 2), spinning mop pads (mop 2), a 42mm step climb (thresholds and climb 1), 26,000 Pa suction and a 250-minute battery (power 2), a HEPA filter and an anti-tangle brush (upkeep savers 1). That is 10 points, and with real mapping nothing caps it: 10 / 10, Tier 4.

A tidy mid-range machine. Camera vSLAM mapping with structured-light obstacle avoidance (navigation and avoidance 1), a self-empty dock but no mop wash (dock 1), a vibrating mop (mop 1), only a 20mm climb (thresholds and climb 0), 15,000 Pa suction and a 180-minute battery (power 1), a HEPA filter but no anti-tangle brush (upkeep savers 0.5). That is 4.5 points: Tier 1. Add an AI camera for proper obstacle avoidance and it reaches 5.0 - Tier 2; swap the camera mapping for LiDAR and it is 5.5; strip the mapping out entirely and the nav gate holds it at Tier 1 no matter the rest.

The live tier list

Here is every robot vacuum we currently track that is in stock, ranked by its Dust Rocket Autonomy Score and grouped into tiers. Each row shows the score, the points earned in each of the six parts, the navigation type, and the cheapest price we are tracking right now. The list re-ranks itself throughout the day as prices and the machines we track change.

Want the shopping version of all this? Our robot vacuum buying guide walks through which features are worth paying for. For how we gather and check prices, see how we source our prices.

Methodology changelog

We version this method openly so the number stays trustworthy as we refine it. The score is always out of 10; the component weights and the tier bands can change, and every change is logged here.

• v3 (1 November 2026) - Navigation became a two-point part. Mapping keeps one point (laser 1, camera or other mapping 0.5, none 0) and obstacle avoidance now earns the second, scored on its own (AI camera or vision 1, structured light or laser 0.5, none 0), so a machine that steers around cables and socks is rewarded for it. Obstacle avoidance moved here out of the old "Terrain and hazards" part, which is now "Thresholds and climb" and worth a single point for step-climb. The tier bands were nudged down 0.5 (Tier 3 from 7.0, Tier 2 from 5.0, Tier 1 from 3.0) to keep the tier populations steady now that avoidance is graded, and the tier labels moved to plain automation levels - Very high, High, Moderate, Basic and Minimal automation - paired with a note that this is a specification-based score. A machine with no obstacle avoidance now shows a hands-off limitation alongside its number. The six parts still sum to a genuine 10 and the mapping gate (no published mapping is capped at Tier 1) is unchanged.
• v2.1 (31 October 2026) - Navigation became a scored component worth 1 point, so the six parts summed to a genuine 10 rather than topping out below it. The tier bands were re-based upward to match the fuller scale, and the component previously called "Maintenance extras" was renamed "Upkeep savers". The nav gate (a machine with no published navigation is capped at Tier 1) was unchanged.