A predictive survey says the aisle sits at -62 dBm. The robot driving down that aisle reports -78 dBm. Performance Networks puts the gap between what predictive surveys report and what autonomous mobile robots (AMRs) actually measure at 12 dB to 20 dB, and that gap is where most warehouse automation projects start to fail.
The failure is not dramatic though. A handheld scanner that loses its association for a few hundred milliseconds simply retries, and the picker never notices. An AMR loses its position update for the same few hundred milliseconds, and the fleet manager does the safe thing: it stops the robot. At a typical travel speed of 1.5 metres per second, the vehicle has already covered most of a metre with no fix. One stopped robot blocks the aisle behind it.
Net result? Throughput falls, and nobody can point at a broken component, because nothing is broken.
The robot is the weakest client on your network
Warehouse Wi-Fi has been designed around scanners for twenty years, and scanners are forgiving. Robots are not.
Most AMRs carry a single-chain 1×1 radio where a 2×2 radio would give real resilience. Their antennas sit inside a protective housing, in a position chosen by a mechanical designer rather than an RF engineer. The payload makes it worse: a metal cage or a loaded pallet blocks, absorbs and reflects signal, so the RF profile of the vehicle changes every time it picks something up, and changes again each time it turns.
Then there is roaming behaviour. Robot vendors write conservative roaming logic on purpose, because they want determinism and they do not want unnecessary state changes mid-journey. Support for the fast-roaming amendments, 802.11k, 802.11v and 802.11r, is patchy. The practical result is a client that clings to a degrading access point (AP) long after a phone would have moved on, and that depends on clean, predictable overlap between cells to hand over at all.
Designing to the scanner specification and hoping the robots cope is the most common and most expensive mistake on these sites.
Roaming is the design target, not coverage
A coverage map is easy to satisfy. Common practice is 95 per cent of floor area at -67 dBm or better, which most integrators can hit with enough access points. Handover is the harder problem, and it is the one automation depends on.
Aim for 15 to 20 per cent cell overlap at the -67 dBm boundary along every route a vehicle takes. Below roughly -75 dBm, sticky client behaviour begins, and a 1×1 robot radio reaches that point sooner than the tablet you validated with. Set a minimum RSSI threshold and disable the low data rates, so that distant clients are pushed to re-associate rather than allowed to limp along at 6 Mbps and consume airtime that the whole cell has to share.
Turn on 802.11k so clients receive neighbour reports instead of scanning blindly, and 802.11r so encryption keys are cached across APs and the client skips a full authentication on every handover. Check the caveats before you commit. On Meraki hardware, for example, 802.11r is not supported in NAT mode, with distributed layer 3 roaming, or with WPA3-Enterprise 192-bit mode. Some robot radios will refuse to associate at all when fast transition is enabled, which is a reason to test with the actual vehicle before go-live.
Use 20 MHz channels in the 5 GHz band. Wider channels look better on a datasheet, but a scanner sends tiny packets and a robot sends small, frequent position updates. Channel reuse across a long building matters far more than headline throughput.
Three device classes, one RF plan
| Device class | Typical radio | Design target | First symptom of a bad design |
|---|---|---|---|
| Handheld scanners, vehicle-mount terminals | 2×2, 5 GHz; some legacy units 2.4 GHz only | -67 dBm across 95 per cent of floor area | Scan retries, slower picks |
| AMRs and automated guided vehicles | Often 1×1, conservative roaming, limited 802.11k/v/r | -65 dBm with 15 to 20 per cent overlap on every travel path | Robot halts, aisle blocks |
| Sensors, RFID readers, VoIP handsets | Low rate, frequently 2.4 GHz | Coverage through the full rack volume, not just the floor | Missing telemetry, dropped calls |
The three classes share one set of channels and one noise floor. That is the constraint the design has to resolve.
What the Ofcom 6 GHz decision changes for UK sites
Ofcom published its final 6 GHz statement on 20 July 2026. The lower band, 5925 to 6425 MHz, is available at standard power under automated frequency coordination. The 6425 to 6585 MHz slice is designated Wi-Fi priority for low power indoor use without coordination, and 6585 to 7125 MHz is mobile priority with coordinated Wi-Fi access. Very low power devices are permitted across both.
For a distribution centre, the honest assessment is that 6 GHz is useful but not yet for the robots. Very few AMRs ship with a 6 GHz radio, and the shorter range of the band suits dense offices better than a 12 metre clear span. Its real value is offloading: move tablets, laptops and newer handhelds to 6 GHz and you free 5 GHz channels for the vehicles that have nowhere else to go. Keep 2.4 GHz alive on channels 1, 6 and 11 only, and only for the legacy scanners that genuinely need it.
Survey the way the robot sees the building
Specialist warehouse wifi designers treat the survey as three separate exercises, and skipping any one of them produces the dBm gap described at the top of this article.
A predictive design sets the AP count and rough positions. An on-site AP-on-a-stick test confirms how the actual racking and stock attenuate signal, which no model gets right. Validation then measures with the real client devices, at the real antenna height. Surveying at 1.5 metres with a laptop tells you nothing useful about a radio mounted at 400 mm inside a steel chassis.
Racking geometry drives the rest. Mount APs at 3 to 4.5 metres on rack end caps or columns and point them down the aisle, rather than hanging them from a 12 metre roof where the inverse square law does the damage before the signal reaches anything. Antenna choice follows mounting height: a 2.2 dBi omnidirectional antenna that works at 6 metres will not serve a high-bay aisle, where a higher gain or directional pattern is needed. Watch the power budget too, since modern APs draw PoE+ at 30 W or PoE++ above 60 W, across cable runs capped at 100 metres.
Survey the building full if you can, then again when it empties. Stock is the single largest variable in the RF environment, and it moves every day.
What to ask before you buy
Wi-Fi 7 multi-link operation is the technology that should eventually solve this, by letting a client hold links on two bands at once and keeping latency predictable through interference. Access points supporting it are shipping now. The robots are not, and a fleet bought in 2026 will still be on the warehouse floor in 2033.
So, the useful question is not which Wi-Fi generation to install. It is about how many spatial streams, which bands, and which of 802.11k, 802.11v and 802.11r does the radio actually support in firmware. Getting a straight answer to that is much better than another six access points.
