HD maps—the hidden sensors that help autonomous vehicles see round corners

With fresh data at its disposal, the HD map keeps automated driving systems well-informed on changes up ahead. By Freddie Holmes

Autonomous vehicles (AVs) will use vision sensors to scrutinise all visible elements of the road, but cameras and laser scanners cannot see around corners. Like any good driver, preparation is key to ensuring the vehicle enters the unknown in a safe manner, slowing down in case of traffic or a spot of black ice. Whereas human drivers have to rely on their senses to prepare, the AV has an ace up its sleeve: high definition (HD) maps.

These maps can help an AV to accurately locate itself on the road and understand what is coming up ahead. The vehicle will know exactly where it is on the road to within decimetres, and is well informed of road conditions miles ahead. While it is not the be-all and end-all, the HD map is considered to be another crucial element to any automated driving system, even low-level forms of driver assistance features.

“We see the map as a complement to the sensors of a vehicle, and some of the use cases are about real-time changes to the road—traffic jams, road works, lane closures and so on,” explained Frans de Rooij, Regional Director Product Management, Autonomous Driving at TomTom. “The live layers are able to warn the AV’s path planning system before the sensors are able to perceive these changes.”

Level up

HD maps are generated from a number of sources, one of which is TomTom’s own fleet of survey vehicles. These cars have been navigating the globe for years now, and are kitted out with a tiara of highly accurate sensors mounted on the roof, just like an autonomous test vehicle.

The primary sensor is the rotating LiDAR, which delivers 700,000 data points per second. These data points indicate the precise location of objects in the surrounding environment. A 360-degree camera and two GPS antennas complete the array, and are paired with other sensors that monitor critical elements of the vehicle, such as the traction control and an accelerometer. Across Europe, the US, Japan and South Korea, this fleet has already produced more than 400,000 kilometres of HD map coverage.

A key motivator has been the industry’s push to democratise highway driving assistance features such as General Motors’ Super Cruise, Tesla Autopilot, Nissan ProPilot and Volvo Pilot Assist. These systems can, under defined parameters, partially automate the vehicle’s steering, acceleration and braking. Drivers are required to remain alert and ready to take over if necessary, and with current systems still in their infancy, disengagements are not rare. TomTom sees an opportunity to improve the capabilities of these features and add an additional degree of redundancy.

“The largest commercial opportunity we see over the next few years is adding HD maps to these highway features, which are already being rolled out in large volumes,” said de Rooij. “Level 2 systems often start out without a map, but gradual increases in functionality often require a HD map to bring the necessary redundancy and allow the driver to take his eyes off the road for a few seconds or longer. There’s a significant commercial opportunity with these highway features, and this is why we’ve created our HD map for all the controlled access highways across the US, Europe, Japan and South Korea.”

AVs need fresh data

As ever, fully autonomous driving systems of Level 4 and above are also of significant interest. The addressable market may be far smaller today, but the overall need for HD maps is far higher. That additional level of redundancy could prove vital if other sensors on the vehicle fail or become obscured by mud, for example. “High levels of automation in urban areas has been well publicised, with robo-taxis already running around in Arizona and Pittsburgh, for example. We’re ready to ramp up when demand takes off, but we see most demand—at least in terms of volume—coming from partially automated highway systems.”

However, HD maps are only useful if the data underpinning them is fresh. Otherwise, the vehicle is essentially relying on the past. A rapid update cycle ensures that the vehicle is aware of changes on the road in real-time, such as a recent crash or a new road sign. “This brings what we call ‘the need for speed’,” said de Rooij. “The map is only relevant if it is up to date; accuracy is great but if you’re accurately showing yesterday’s road then it’s useless.”

Updates are only pushed out to the map once sufficient data has been collected to accurately confirm any changes that need to be made. This requires “quite a bit of technology,” continued de Rooij. A transactional map-making platform on the back-end means that TomTom can process the data and perform quality control before releasing each individual change. TomTom’s AutoStream technology can then deliver updates in small ‘tiles’ to the vehicle cache. “You need to get this right because it ensures quality and consistency with the rest of the map,” affirmed de Rooij.

Let’s collaborate

While TomTom has significant scale at its disposal, it has entered numerous joint development programmes with other leading players in the AV space. In addition to a long-standing partnership with Bosch, other notable partners include Baidu, Zenuity, Nvidia and Qualcomm. A collaboration with Japanese Tier 1 supper Denso was announced at CES 2019, which will see the pair develop a Level 2 system designed for highway driving and main roads.

“It is important to stress the need for collaboration,” concluded de Rooij. “We make maps for autonomous driving but that’s only one aspect; you need more components to make a full system. We work with different sensor developers—be it an accurate GPS, camera, LiDAR or radar—to ensure our map has the right data to correlate into the map. We are working very closely with the relevant parties to create a complete system.”