AI Self Driving

The DeepMap Approach: “Your Data, Your HD Map”

In a near future, when self-driving cars have replaced human drivers, it's believed that high-definition (HD) maps can play a similarly useful role, as “the memory of robot drivers”.

Human drivers can navigate better in areas where they’ve been before, as memories of the previous experiences can help inform their understanding of the environment. In a near future, when self-driving cars have replaced human drivers, it’s believed that high-definition (HD) maps can play a similarly useful role, as “the memory of robot drivers”.

Palo Alto-based startup DeepMap wants to accelerate the creation of such useful memories. Founded in 2016 by a team of map veterans, the company provides affordable HD mapping and localization services for production-level autonomous vehicles (AVs).

Earlier this month Generation Investment Management joined DeepMap’s high-profile investor list that includes Nvidia, Bosch, Andreessen Horowitz, and many others. The autonomous driving startup finds itself in the enviable position of being courted by both automotive leaders and tech giants. DeepMap’s recent fundraising has pushed its valuation to US$450 million, according to a TechCrunch report.

The reason for all the interest is simple: the HD map industry will climb to an estimated market value of US$9.4 billion by 2025, according to Goldman Sachs.

HD maps are a 3D representation of real-time road network data, including lanes, road boundaries, objects, etc. Robot drivers are not smart enough to make driving decisions from traditional maps, but they can easily and precisely localize their own and other vehicles’ positions on the road in real time using HD maps.

HD Map

Major players such as Alphabet’s self-driving unit Waymo and GM’s Cruise already have their autonomous test vehicles on roads collecting data and building HD maps; as do dedicated map producers such as Here Technologies and TomTom. Meanwhile, startups that cannot afford the extremely high production costs of mapping platforms like Mapbox and IvI5 are effectively crowdsourcing data from app users.

Rather than producing one HD map database and hoping it works for all AV solutions, DeepMap provides a set of services, including hardware tools, software solutions and field data collection functions that can transform customers’ self-driving fleet data into their own personalized HD maps.

DeepMap Co-founder and CEO James Wu told Synced that HD maps are becoming a tightly focused and independent industry sector. Prior to founding DeepMap, Wu led the creation of maps and architectures at Google, Apple, and Baidu. He came to the conclusion that autonomous driving is too complex for a single vendor to provide a complete full-stack solution. HD maps are the fundamental part of such a stack.

“HD maps touch upon every component in autonomous driving, ranging from vehicle sensors, in-vehicle computing and communications, data collection and sorting, fleet management, data processing integration, map production maintenance, and labeling… it is a highly complicated system engineering problem,” says Wu.

HD maps remains a nascent industry at the R&D stage, wrestling with several challenges. Centimeter-level precision is a must: “The dividing line between lanes is only ten centimeters wide. If you are a few centimeters off, two cars might collide, or a car might even fall off a cliff,” cautions Wu.

Other challenges include refreshing HD maps frequently while not compromising their precision, scaling HD maps from a very limited testing range to a much broader scope, and factoring in weather and other dynamic environmental considerations.

DeepMap’s customers in America, Europe, and East Asia are already using its service to generate HD maps. This is, however, a cash-burning and time-consuming business. Google spent almost ten years on their map system, and developing a comprehensive HD map system is expected to take much longer.

Building and maintaining high-precision HD maps within a reasonable budget is a problem facing all HD map makers. Chinese media has reported that a single data-collecting autonomous vehicle from leading Chinese map maker AutoNavi Maps costs more than US$1 million.

“The world’s largest mapping companies are not making money. Think about it, the terabyte-level data collected by self-driving cars each day causes huge loses to the data processing machine. Machine power, labor and software development also add to the huge cost,” says Wu.

Wu is confident that the high production costs will fall over the next five years as global demand for HD maps surges. Even if the full democratization of self-driving vehicles does not happen in that span, vehicles with advanced driver-assistance systems (ADAS) will also have an increasing reliance on HD maps and drive the market.

Building on its strong partnerships and investments, DeepMap is now expanding its presence in China. The company has a Beijing branch office, and is currently seeking local talents for R&D. To perform data collection in China, foreign companies would likely team up with a local firm to obtain the special authorization from local administrative departments required for processes such as surveying and mapping.

DeepMap does not rule out the possibility of building their own HD maps, but Wu sees his company as more of an enabler. “In the long run, our philosophy is ‘Your data, your map.’”

Journalist: Tony Peng | Editor: Michael Sarazen


1 comment on “The DeepMap Approach: “Your Data, Your HD Map”

  1. Very interesting post. This is my first time visit here. I found so many interesting stuff in your blog especially its discussion. Thanks for the post!

Leave a Reply to ferahtia.FS Cancel reply

Your email address will not be published. Required fields are marked *

%d bloggers like this: