RTLS accuracy — what the numbers actually mean.

Per-technology realistic accuracy under production load.

UWB (TDoA, well-designed): 10–30 cm @ 95th percentile controlled; 30–60 cm in a typical industrial bay; degrades to 100+ cm if anchor density is wrong.

BLE-AoA: 50 cm to 2 m @ 95th percentile depending on Locator density, antenna geometry and ceiling height.

RAIN RFID (location-grade): 1–3 m at item-level under a ceiling phased-array; presence at portal read points.

GPS / GNSS (standard): 3–5 m outdoor; does not work indoors.

GNSS-RTK: 1–3 cm outdoor with corrections; requires unobstructed sky view.

LoRaWAN geolocation: 100–500 m via TDoA across multiple gateways; good for presence, not position-precise.

SLAM (visual / lidar): 5–20 cm in mapped environments; degrades sharply if the environment changes.

Why vendor accuracy claims are usually 95th percentile under lab conditions.

Most spec-sheet numbers assume line-of-sight, designed-density placement, calibrated mounting, no production-load interference, single-tag testing.

Real production environments degrade those numbers by 30–100%.

Ask vendors to quote 95th-percentile accuracy under your environment. They will resist; that resistance is the answer.

How accuracy degrades — the five common causes.

Multipath from metal racking, glass, stainless panels, concrete-aggregate floors, HVAC ducting.

Anchor mounting height drift — 200 mm off design changes UWB cone of coverage by 15–25%.

Synchronisation drift — TDoA requires nanosecond sync; mis-cabled PoE switch quietly destroys accuracy.

RF blockers introduced post-survey — racking, guarding, mezzanines.

Production-load interference — forklift mast occlusion, dock-door cycles, press cycles.

How to specify accuracy so vendors cannot hide behind lab numbers

RTLS accuracy is meaningless without percentile, environment and use-case conditions. A vendor quoting “10 cm” usually means a best-case or median result in a clean lab. Operators should specify accuracy as: technology + percentile (50th and 95th) + environment class (open / partitioned / metal-heavy) + tag motion + update rate + whether first-path NLOS was present. That sentence is what belongs in RFPs and pilot gates.

Realistic production bands. UWB: often 10–30 cm when geometry is good; can stretch toward 0.5–1.5 m under industrial NLOS if anchors are poorly placed. BLE AoA: roughly 0.5–3 m depending on density and multipath; BLE RSSI alone is typically multi-metre zone-level. Wi-Fi RTT: often 1–5 m depending on AP density. Passive RFID portals: not continuous accuracy — they deliver presence at a read point with a read-rate KPI instead.

Match accuracy to the locating decision. Warehouse asset find: 1–3 m may be enough. Ward equipment: room-level. WIP station presence or OR table distinction: sub-metre. Forklift–pedestrian safety: centimetre-class with low false negatives. Paying for UWB everywhere to answer room-level questions is a common budget failure.

Why pilots look better than day 60. Empty bays, ideal tag mounts, low traffic and vendor-tuned filters inflate early numbers. Production load adds mast occlusion, moving metal, body shadowing and layout drift. Build acceptance tests for peak conditions and re-measure after the first layout change.

Contracting tip. SLAs that only guarantee accuracy under the installer’s commissioning conditions are weak. Tie acceptance to your operational KPI and percentile under agreed live-load scenarios, and keep a remediation path (retune density/orientation) before any automatic hardware uplift.

Reporting accuracy to boards and vendors without losing the plot

Report locating accuracy as distributions over time and zone classes. A single site-wide average hides the cell that fails safety thresholds and the ward that already meets room-level needs. Pair accuracy charts with operational KPIs so leadership sees whether centimetres (or metres) actually changed search time, dwell or utilisation.

When vendors present new firmware claims, re-run the same production-load script before accepting a contractual accuracy uplift. Firmware can help; it does not repeal multipath. Keep a living accuracy register per zone as part of managed operations — especially after layout changes.

For hybrid estates, never average UWB and BLE error into one number. Publish separate bands per radio layer so buyers and operators set expectations correctly.

Accuracy and latency are twin locating requirements

A precise fix that arrives too late cannot drive a safety interlock or a station confirmation. Specify latency alongside accuracy in every pilot gate. Some workflows tolerate multi-second BLE updates; others need sub-second UWB. Measuring only centimetres while ignoring time-to-event produces systems that look accurate in reports and fail in operations.

How to specify accuracy so vendors cannot hide behind lab numbers

RTLS accuracy is meaningless without percentile, environment and use-case conditions. A vendor quoting “10 cm” usually means a best-case or median result in a clean lab. Operators should specify accuracy as: technology + percentile (50th and 95th) + environment class (open / partitioned / metal-heavy) + tag motion + update rate + whether first-path NLOS was present. That sentence is what belongs in RFPs and pilot gates.

Realistic production bands. UWB: often 10–30 cm when geometry is good; can stretch toward 0.5–1.5 m under industrial NLOS if anchors are poorly placed. BLE AoA: roughly 0.5–3 m depending on density and multipath; BLE RSSI alone is typically multi-metre zone-level. Wi-Fi RTT: often 1–5 m depending on AP density. Passive RFID portals: not continuous accuracy — they deliver presence at a read point with a read-rate KPI instead.

Match accuracy to the locating decision. Warehouse asset find: 1–3 m may be enough. Ward equipment: room-level. WIP station presence or OR table distinction: sub-metre. Forklift–pedestrian safety: centimetre-class with low false negatives. Paying for UWB everywhere to answer room-level questions is a common budget failure.

Why pilots look better than day 60. Empty bays, ideal tag mounts, low traffic and vendor-tuned filters inflate early numbers. Production load adds mast occlusion, moving metal, body shadowing and layout drift. Build acceptance tests for peak conditions and re-measure after the first layout change.

Contracting tip. SLAs that only guarantee accuracy under the installer’s commissioning conditions are weak. Tie acceptance to your operational KPI and percentile under agreed live-load scenarios, and keep a remediation path (retune density/orientation) before any automatic hardware uplift.

Reporting accuracy to boards and vendors without losing the plot

Report locating accuracy as distributions over time and zone classes. A single site-wide average hides the cell that fails safety thresholds and the ward that already meets room-level needs. Pair accuracy charts with operational KPIs so leadership sees whether centimetres (or metres) actually changed search time, dwell or utilisation.

When vendors present new firmware claims, re-run the same production-load script before accepting a contractual accuracy uplift. Firmware can help; it does not repeal multipath. Keep a living accuracy register per zone as part of managed operations — especially after layout changes.

For hybrid estates, never average UWB and BLE error into one number. Publish separate bands per radio layer so buyers and operators set expectations correctly.

Accuracy and latency are twin locating requirements

A precise fix that arrives too late cannot drive a safety interlock or a station confirmation. Specify latency alongside accuracy in every pilot gate. Some workflows tolerate multi-second BLE updates; others need sub-second UWB. Measuring only centimetres while ignoring time-to-event produces systems that look accurate in reports and fail in operations.