Many vehicle owners are familiar with automated safety features that fall under the umbrella term of Advanced Driver-Assistance Systems (ADAS). Examples of ADAS features available today include:
- Adaptive cruise control,
- Automatic braking,
- Collision avoidance systems,
- Lane-change assistance, and more.
Automakers are now in a fierce competition to offer the next generation of ADAS functionality. These systems are rolling out in new car and truck models now and over the next few years, offering features such as:
- Automatic lane change over multiple lanes,
- Hands-off driving from entrance-to-exit,
- Advanced parking assistance, and more.
In the automotive industry, this market space is known as “Level 2+,” a term popularized by Mobileye and NVIDIA to define the technology gap between SAE Level 2 and Level 3 autonomous driving.
Others refer to it as Autopilot (Tesla), Propilot (Nissan), and SuperCruise (GM). It is also described simply as “hands-off” driving or the “new ADAS.”
Analyst Phil Magney of VSI Labs explains it this way: “Level 2+ is the new ADAS and is on every automotive OEM’s roadmap. At least 15 OEMs either started producing or are planning to come up with Level 2+ ADAS systems this year and next year.”
The sense of urgency to design new ADAS systems into new car models intensified during 2020 for two main reasons: a) significant advances in sensors, software, and assisted-driving technologies; and b) worldwide attention on Tesla’s Autopilot Full Self-Driving (FSD) beta. (Note: For a detailed look at FSD, see Brad Templeton’s excellent analysis in Forbes.)
The new ADAS market is expected to be huge. A report by Expert Market Research titled Global Advanced Driver Assistance Systems (ADAS) Market Report and Forecast 2020-2025 predicts a 2025 ADAS market size of $65 billion (USD).
A hot topic when discussing new ADAS features – such as automatic lane changing – is the role that highly-detailed maps (that is, maps designed to be read by machines, not humans) will play at the core of L2+/L3 systems.
In this post, I will make the case as to why automakers would be remiss to not utilize maps for next-generation ADAS systems. In fact, I propose that the most significant difference between today’s ADAS and the new ADAS will be the use of maps to extend the functional capabilities and safety – as well as comfort – of L2+/L3 systems.
The better the map, the better the system
New ADAS driving systems are designed to drive a vehicle with minimal input from the human driver, while still requiring driver supervision in a variety of situations and conditions.
In general, these driving systems use some sort of a machine-readable map to help them do their job safely and comfortably. Different systems vary in how accurate and detailed the maps are, what is mapped, and how the maps are kept fresh. The bottom line is, the better the maps, the better the system. Here’s why:
- Maps let the vehicle determine its current location better.
- Maps let the vehicle establish lane geometry accurately and reliably in order to position the car in the correct lane.
- Maps increase driving comfort by looking ahead and optimizing the driving path.
- Maps see around blind corners and can predict which lane to drive in.
- Maps know where the lanes are even when lane lines are missing or worn out.
- Maps let the system know about hazards and potentially dangerous areas to improve comfort and safety.
- Maps know where to park the car safely in a dangerous situation.
- Maps have a solid and verified understanding of the rules of the road and contain road controls like traffic signals and signs for safe and legally correct driving.
- Maps can inform the vehicle about expected accuracy and confidence levels for the sensors actively used in the car to guarantee safety.
What’s in the map?
What’s in the map will vary with the type of functionality and features the new ADAS system is enabling. Typically, the following elements may be contained in the map:
- Navigation and routing capabilities based on lane geometry of roads (often found in advanced navigation systems today).
- Precise lane geometry with the shape and location of all lane borders, and traffic rules for that lane (lane topology).
- Location and meaning of any traffic controls (signals and signs) as well as parking zones and parking rules, pickup zones, or other places with specialized rules.
- Location of curbs, barriers, potholes, speed-bumps, and other hazards.
- Full 3D geometry of roads (hills, ramps, and slopes).
- Perception aids, such as static radar targets.
- Localization clues, which help the vehicle place itself on the map, including:
- 2D images of the road surface, cracks, and features, and positions, shape, and flaws of lane markers.
- 3D positions of elevated objects — curbs, signs, trees, boulders, poles, lamps, buildings, etc.
What can maps do for the new ADAS?
With all of the features, capabilities, and considerations described above, what can a map do to make hands-off driving safer, more comfortable, and more efficient?
An L2+/L3 system does not have to be as “near perfect” (in terms of accuracy, completeness, and reliability) as a fully autonomous self-driving system (L4/L5), but the better and more complete it is, the more relaxing and safer the experience will be for the drivers and passengers, as well as for other vehicles on the road.
There are different new ADAS systems on the market today and there are many more to come. What are the biggest differentiators between these systems as far as maps are concerned? Here are some of them:
- The way the maps are created and updated,
- The mapping approach and the map quality (accuracy and reliability),
- The sensor sets used and their specific accuracies to create and update the map (LiDAR or not, camera configuration, GNSS, radar, and so on),
- The quality of the perception output, which can be used to update the map and keep it fresh, and
- The planning quality after localization, map, and perception are completed.
While it is possible to drive hands-off without a map, or with poor maps, an L2+/L3 system will make fewer mistakes if it has a high-quality map to provide extra information. Fewer mistakes will result in:
- Fewer times the ADAS system will have to disengage due to an ambiguous or difficult situation the system can not handle with the required safety and reliability.
- Fewer times the car drifts out of a good driving line. Drifting may cause drivers to intervene, and with each intervention, drivers and passengers lose trust.
- Fewer accidents, including roll-overs, rear-end collisions, and head-on collisions.
Maps do more than help a car to position itself properly in the lane. Maps help the car to understand the meaning of the road and its surroundings, not just the geometry.
Maps represent the collectively accumulated and verified information about the road geometry and the lane topology, as well as all safety and traffic relevant information to drive legally and safely. Maps can expand the geographical area an ADAS system operates in by adding information and safety features about new roads.
Maps can avoid false-positive detections (detecting something which is not there), which would cause the system to respond in an unnecessary way.
More importantly, maps can avoid false negatives (being blind to something that is actually there), situations where a system might not see something, but the map has the information about it and the system can respond in time. And, maps can handle different types of roads and weather conditions, with confidence in making fewer mistakes.
Summary
Although L2+ vehicles can be operated without a map, the key safety risk that a high-quality up-to-date map can mitigate is the following:
- In theory, a driver in an L2+/L3 vehicle is responsible for watching the road and taking control whenever necessary.
- In reality, drivers are not always ready to take over promptly.
This is where maps excel. Map-based systems reduce mistakes and ambiguities, leading to a better driving experience, an expanded operational design domain (ODD), and most importantly, enhanced safety and comfort.
At the NVIDIA GPU Tech Conference in October 2020, DeepMap announced DeepMap HDR™ Map and Map Update for automakers and suppliers, tailored to Level 2+/3 driving systems. (See Future-Proof Mapping for Level 2+ Autonomy and Beyond by DeepMap co-founder and CTO Mark Wheeler.) 
DeepMap HDR services offer the ability to combine high quality data from dedicated survey vehicles to create a highly accurate base map and to update the maps with crowdsourced data from an array of perception sensors like camera, radar, LiDAR, and other sensor sets.
DeepMap HDR services are designed to deliver the best possible HD maps for L2+/L3 driving for an affordable price and with the required scalability. It solves the geographical and functional expansion of the ODD and the map update with crowdsourced data.
We’d appreciate the opportunity to discuss this post and DeepMap’s future-proof Level 2+ through Level 4 mapping services with you. You can see highlights of recent DeepMap mapping projects here.
Our contact details are:
Partnerships: [email protected]
Marketing: [email protected]
Business Development: [email protected]
About DeepMap
DeepMap is accelerating safe autonomy by providing the world’s best autonomous mapping and localization solutions. DeepMap delivers the technology necessary for self-driving vehicles to navigate in a complex and unpredictable environment. The company addresses three important elements: precise high-definition (HD) mapping, ultra-accurate real-time localization, and the server-side infrastructure to support massive global scaling. DeepMap was founded in 2016 and is headquartered in Palo Alto, Calif., with offices in Beijing and Guangzhou, China. Investors include Andreessen Horowitz, Accel, GSR Ventures, Generation, Goldman Sachs, NVIDIA, and Robert Bosch Venture Capital. For more information, see www.deepmap.ai.
