HD Mapping
HD mapping is the creation of extremely precise digital road maps in which lane edges, signs, and traffic lights are recorded to within centimeters. Self-driving cars use such maps to know what to expect at a given location before their sensors actually see it.
A normal navigation map on a phone shows streets as rough lines. That’s enough to guide a person along the way. A car that steers without a driver needs considerably more. It must know exactly where the lane marking runs, how high the curb is, and at what height the traffic light hangs. HD mapping is the creation of exactly such maps. The abbreviation HD stands for High Definition, meaning high resolution: positions are recorded to within a few centimeters instead of several meters.
Why cars need a memory of the road
The cameras and lasers of a self-driving car only ever see the present moment. In heavy rain, low sun, or snow on the road, they see even less. An HD map acts here like a memory. The car knows that an intersection follows after the curve, even if it cannot yet perceive it.
The second benefit is computational load. Without a map, the software would have to reinterpret every detail of the surroundings anew. With a map, it essentially only has to check whether reality matches the expectation. That is faster and less error-prone. This is exactly why companies like Waymo rely heavily on pre-built maps for their robotaxis.
But there is also an opposing position. Tesla largely forgoes HD maps and wants to drive using camera images alone. The argument behind this: maps become outdated, and a system that relies on them fails at a new construction site. Which approach will prevail remains an open question to this day.
From survey drive to centimeter-precise map
It usually starts with a survey vehicle. Mounted on the roof is a lidar, a sensor that measures distances using laser beams. It scans the surroundings millions of times per second. Added to this are several cameras and a very precise satellite receiver. From these measurements, a point cloud is initially generated, a cloud made up of millions of individual measurement points in space.
This point cloud is not yet a map. In the next step, software assigns meaning to the points. It recognizes which points belong to the roadway, which to a traffic sign, and which to a tree. Humans then review the results and correct errors. The end result is a layered model: lanes, traffic rules, fixed objects.
The most labor-intensive part is not the creation but keeping it up to date. Roads get rebuilt, markings disappear, traffic lights get added. That’s why many vehicles report deviations back to the map provider during operation. This is called crowdsourcing: many cars deliver small observations, which together produce an update.
Who builds HD maps and where they are already driving
In business news, HD mapping mainly appears via company names. Here Technologies is owned by several German carmakers, TomTom is known from navigation devices, and the Israeli company Mobileye belongs to Intel. All of them sell map data to automakers. Such data is a business with recurring revenue, because the updating never ends.
The technology becomes visible wherever robotaxis are on the road, for instance in Phoenix, San Francisco, or Wuhan. It is notable that these services initially operate only in clearly defined city areas. One reason for this is precisely the map problem: every new district must first be surveyed.
A common misconception is that HD mapping is just a more precise version of Google Maps. The difference lies in the purpose. A navigation map answers the question of which route to take. An HD map answers the question of exactly where, down to the centimeter, the vehicle currently is and what applies directly ahead of it. Even modern driver-assistance systems in production cars already use scaled-down forms of this, for instance for lane-keeping on highways.