The GPS in your phone is accurate to somewhere between 3 and 10 metres under good conditions — more than enough to find the right street, hopelessly imprecise for knowing which of three lanes a car is actually in. An automated driving system built on that level of accuracy would be guessing whether it’s centred in its lane or drifting into the next one. Getting from “somewhere on this road” to a confidently known position accurate to a few centimetres, updated many times a second, took stacking several separate technologies on top of each other, none of which is sufficient on its own.
Why plain GPS was never going to be enough
A standard GNSS (Global Navigation Satellite System) receiver — GPS being the best-known of several such systems, alongside Europe’s own Galileo, Russia’s GLONASS and China’s BeiDou — calculates position by timing radio signals from satellites, and that calculation carries inherent errors from atmospheric interference, satellite geometry and signal reflections. In a genuine urban canyon, tall buildings on both sides of a street reflect satellite signals before they reach the receiver, a problem called multipath that can throw the calculated position off by tens of metres exactly in the dense city environments where precise positioning matters most. Tunnels and covered car parks are worse still: no satellite signal reaches the receiver at all.
Correction signals: turning metres of error into centimetres
The first fix is augmented GNSS: a network of fixed ground reference stations at precisely known locations continuously compares its own calculated GPS position against its real, surveyed position, and broadcasts the resulting correction data to nearby receivers in real time. Two competing techniques do this: RTK (Real-Time Kinematic) compares raw satellite signal phase between a nearby base station and the vehicle for very fast, very precise corrections, typically requiring a reference station within a few tens of kilometres; PPP (Precise Point Positioning) uses a sparser global network of reference stations and highly accurate satellite orbit and clock data instead, trading a slightly slower convergence time for coverage that doesn’t depend on a nearby local station. Either way, the output is the same: GNSS error shrinks from metres down to single-digit centimetres, as long as the satellite signal is available at all.
Filling the gaps: the inertial measurement unit
For the moments GNSS signal disappears entirely — a tunnel, an underground car park, a few seconds of severe multipath — the car falls back on dead reckoning using an IMU (Inertial Measurement Unit): accelerometers and gyroscopes that track how the car’s speed and orientation change from one instant to the next, extrapolating position forward from the last known good GNSS fix. An IMU alone drifts, accumulating error the longer it runs without a fresh satellite fix to correct it, which is exactly why it’s used as a short-term bridge between GNSS updates rather than a standalone positioning system.
What actually makes a map “HD”
Even centimetre-accurate GNSS only tells the car where it is on Earth — it says nothing about which lane that position falls in, where the kerb actually is, or where a specific traffic light is mounted. That’s the job of the HD (High Definition) map, a fundamentally different product from the navigation map in a normal sat-nav: rather than street centrelines and turn instructions, an HD map encodes individual lane boundaries, kerb heights, lane markings, sign and traffic-light positions and road surface geometry, all georeferenced to centimetre accuracy and refreshed far more frequently than a conventional map, precisely because a road repainted or a new sign installed can matter to an automated system in a way it never mattered to a human glancing at a phone screen.
Map matching: snapping the car onto the map it already trusts
With an HD map loaded and a GNSS+IMU position estimate in hand, the car’s onboard sensors — camera and LiDAR, the same sensing hardware covered in our Sense article — compare what they actually see (lane markings, kerbs, road signs, distinctive fixed landmarks) against what the HD map says should be there at that estimated position. This process, called map matching, corrects the GNSS+IMU estimate against known, surveyed ground truth, snapping a merely “very good” position estimate onto a lane-accurate one — the step that actually delivers the centimetre-level confidence the rest of the driving system needs.
Who actually builds an HD map, and how it stays current
Building an HD map at scale is its own specialised industry: dedicated survey vehicles, bristling with LiDAR, cameras and precision GNSS, drive every road that needs mapping and process the resulting sensor data into the layered format described above. Keeping that map current at any meaningful scale, though, increasingly relies on crowdsourced updates — production vehicles already on the road, equipped with the right sensors, continuously comparing what they see against the existing map and flagging discrepancies (a new lane marking, a relocated sign, a road closure) back to the map provider, turning every properly equipped car on the road into a part-time survey vehicle.
The EU research behind building and using these maps
Two complementary EU-funded H2020 projects, both under the Galileo/GNSS downstream applications programme, target opposite ends of the HD mapping problem. INLANE (“Low Cost GNSS and Computer Vision Fusion for Accurate Lane Level Navigation and Enhanced Automatic Map Generation”, coordinated by Vicomtech in Spain, €2.64M, 2016-2018) tackled the car’s side of the problem directly: fusing low-cost GNSS with computer vision to achieve lane-level navigation accuracy without expensive survey-grade hardware, while also automatically generating map data as a byproduct of driving. GAMMS (“Galileo/GNSS-based Autonomous Mobile Mapping System”, coordinated by Geonumerics SL in Spain, €1.38M, 2021-2025) attacked the supply side instead: replacing the traditional 2-3 person survey crews behind conventional mobile mapping systems with fleets of low-cost, autonomous, electrically-powered mapping vehicles built specifically around Europe’s own Galileo constellation, aimed at making HD map creation itself cheap and scalable enough to keep pace with a road network that never stops changing.
Photo: © Car-Shooters