3D Map Making Challenges for Autonomous Driving

Engineering AI

Building accurate road maps is a central part of the effort to build and deploy more autonomous vehicles in the real world. The term “map” may be a bit of a misnomer, though, because these maps aren’t anything like the flat 2D images available online, they’re complete three-dimensional recreations of roadside environments that are updated on a continuous basis to provide a high degree of accuracy — often down to the centimeter scale. These 3D digital maps are a critical part of an autonomous vehicle’s ability to perceive the world, and have key applications in other technologies, which has made the effort to develop the definitive map a highly competitive endeavour.

It’s not surprising that many of the same companies that lead the charge to autonomous vehicles have also taken leading positions in the race toward an effective mapping system. Alphabet, for example, has drawn on its well-developed mapping technology to develop the Waymo autonomous vehicle platform. With Google’s Google Maps, Google Earth, Google Street View, and the navigation app Waze to draw on as reference and inspiration, Alphabet is considered to have some of the best available mapping technology currently available. Tesla, another large company that has made ambitious claims about its self-driving technology, has its own well-developed mapping system. But these large companies represent just one portion of the story, and a very active community of startups is injecting the quest for better maps with creativity and enthusiasm.

This includes Lvl5, a startup founded by Tesla engineers Andrew Kouri, Erik Reed, and iRobot engineer George Tall. To develop more accurate maps, they’ve given San Francisco’s cab drivers a cell phone mount so they can use their phone cameras to take pictures of the road. With the help of more than 2,500 Uber and Lyft drivers, and a computer vision algorithm, the company has mapped over 500,000 miles of roads so far. Lvl5 pays its driver $0.05 per mile — $0.02 for a route that’s already mapped — to take a picture of the road every meter and record the data using the company’s Payver app. Because the Lvl5 solution relies on taxi drivers who drive long distances every day, it claims to be updated quicker and more continuously than its competitors, without requiring the large capital costs associated with building a proprietary fleet of mapping vehicles.

Another start-up to attract great attention is HERE. HERE claims to have developed “self-healing” and “self-forming” map content formed from data it crowdsources via sensors on numerous OEM sources. HERE’s mapping solution creates both a standard 2D representation of the world - which is already popular in the European market, and a 3D live map specifically designed for autonomous vehicles, which recreates the details of the road to an accuracy of 10cm. The company collects 5 billion new records per month through its network of technology partners, while its “HERE Cars” have covered over 3 million kilometers in over 50 countries. HERE, which was acquired from Nokia by Audi, BMW, and Mercedes for a little over $3 billion dollars in 2016, has since formed partnerships throughout the autonomous vehicle industry. A recent partnership with LG promises to merge HERE’s mapping technology with LG’s telematics solution to use the upcoming 5G standard to transmit data from autonomous car sensors, allowing HERE equipped cars to proactively avoid road hazards in real time, including weather and traffic accidents.

But this rush to develop reliable 3D maps may have some unintended consequences as well. This includes a pernicious level of inefficiency, as each and every company maps the same city roads over and over again for their own proprietary solution. An example of this redundancy can be seen in Las Vegas, the host of the annual CES technology trade show, where every year a slew of cars equipped with the latest mapping technologies creates a detailed representation of the city’s streets, a practice that has been going on for a while now. There are other questions to be answered as well. How will we map rural areas, which although less traveled are actually the site of more auto accidents than urban roads? The machine learning tools currently used to label maps are far from perfect as well, meaning these maps still require human intervention to label unrecognised roadside objects manually. How can we make these process more efficient?

These challenges aside, there is no shortage of companies trying to become the dominant map-maker of the autonomous vehicle era. DeepMap, a company formed by veteran engineers from Google and Apple, just recently attracted $25 million dollars of investment, while Amsterdam’s TomTom, a veteran of the navigation and location-services industry, has formed a partnership with Bosch to develop a radar-based road map solution. Although though most analysts agree that ultimately just a few of these companies will survive the tooth-and-nail competition, being the victor of the 3D map competition is just too big an opportunity to deter any of them from trying.

Written by Nick Dale

See Nick Dale's blog

Nick is Senior Director at Verne Global and leads our work across HPC and specifically its implementation within discrete and process manufacturing. He is based in our London headquarters.

Related blogs

Iceland provides the power behind Germany's most pioneering AI start-ups

This week has seen the announcement of Analytic Engineering, a pioneering German AI engineering firm, choosing Verne Global’s data center in Iceland as the location for their intensive computing. This represents another impressive AI and Machine Learning client win for us, following DeepL joining us just before Christmas.

Read more

AI Icelandic Style – Deep Learning in Reykjavik

Last week I was privileged to be part of our AI and HPC Field Trip to Iceland. The goal of the trip was to share insight and observations around the evolution of AI deep neural network (DNN) training and to tour our HPC-optimised data center. The attendees spanned DNN veterans like Eric Sigler of OpenAI, large enterprise data science leaders like Pardeep Bassi of Liverpool Victoria Insurance (LVE) and start-up pioneers like Max Gerer of e-bot7.

Read more

Industrial HPC Solutions: Visualisation

When you imagine what visualisation is in the world of HPC, most people think of astronomy, such as images of galaxies or black holes, or they think of weather, like analyses of tornadoes or hurricanes. Astronomical and atmospheric data is huge, requires HPC to analyse, and can make for amazing, sophisticated visualisations.

Read more

We use cookies to ensure we give you the best experience on our website, to analyse our website traffic, and to understand where our visitors are coming from. By browsing our website, you consent to our use of cookies and other tracking technologies. Read our Privacy Policy for more information.