Robot mowers got good at knowing where they are, not at being left alone

The Verge ran five robot mowers against each other over a summer on a flat three-quarter-acre property in coastal South Carolina, with sandy soil, centipede grass and a lot of trees, and reported that the machines stopped getting stuck ^[1][2]. That moves the competitive question in this category from cutting to localisation, and it leaves one product claim unsupported: the reviewer's conclusion is that these are still not set-it-and-forget-it machines ^[3].

The technical change is narrow and specific. Most automowers dropped boundary wires for RTK, real-time kinematic positioning, which combines GPS with fixed ground reference antennas to place the machine within a centimetre or two ^[4]. The catch was always installation: every mower in the test except the lidar-based Dreame ships with its own RTK antenna that needs a clear view of the sky, and trees, a neighbour's house or your own roofline can block it ^[5]. In earlier testing on the same property, according to The Verge, not one RTK mower ran successfully in the backyard because GPS coverage there was too spotty ^[6]. This year's crop all offer NetRTK, which uses a network of existing reference stations over a cloud connection, and most worked without the reviewer installing an antenna at all ^[7]. When the signal drops, the mowers fall back to lidar or vSLAM ^[8].

Read that as a dependency transfer rather than a fix. NetRTK requires Wi-Fi or 4G on the mower, and coverage varies by address, so the buyer has to check before purchase ^[9]. Every manufacturer currently advertises NetRTK coverage as free for the lifetime of the device ^[10], which is a pricing decision, not a physical property of the service.

The supervision requirement is where the marketing gap sits. One mower spent three days methodically destroying a patch of lawn before the reviewer noticed ^[11]. Another cut through two outdoor extension cords ^[12]. Several dug sandpits in bare patches, and one tried to climb a small tree ^[13]. Obstacle detection improved across all units but still struggles with hoses, cables and cars, to the point where the household now checks under the wheels before driving away ^[14]. Nobody has solved gates, so full-property coverage still needs a human to open one ^[15]. A machine that can quietly run daily and shave millimetres at a time ^[16] is also a machine that can quietly do damage daily, and the three-day detection lag is the number to design around.

The four flagship models tested were priced at $2,499 for the Navimow X430, $2,799 for the Mammotion Luba 3 AWD 3000, $3,099.99 for the Dreame A3 AWD Pro 2500 and $3,299 for the Husqvarna 420 IQ ^[17], an average of about $2,924 ^[18] across an $800 spread ^[19]. At that price the buyer will assume set-and-forget unless told otherwise, and the manufacturer's incentive is not to tell them. The cheaper posture is to sell oversight-assisted autonomy explicitly: a stated check-in cadence, an alert when the machine has been in one area too long, and a coverage precheck at the point of sale rather than after delivery.

Worth watching: whether free-for-life NetRTK survives contact with cloud costs, whether obstacle detection reaches cables and hoses, and whether anyone ships a gate. The reviewer also flags open questions on long-term durability ^[20], which is the metric that decides whether a $3,000 mower is cheaper than the riding mower now sitting idle in the shed ^[21].