Skip to content

Autonomy without HD maps raises serious safety questions

HD Maps are clearly very useful, but are they a prerequisite for autonomous drive capability? By Megan Lampinen

High-definition (HD) real-time mapping is emerging as a pivotal tool in autonomous driving systems. Harnessing input from camera, radar and LiDAR, these maps help automated vehicles to  understand their location and to plan beyond what the sensors themselves can see. But they are also difficult, time-consuming and expensive to create and maintain. Just how important are these HD maps to realising an autonomous future?

“Almost everyone in this business is using maps to varying degrees as part of the control system for their automated driving systems, with a few exceptions,” observed Sam Abuelsamid, Principal Research Analyst at Navigant Research. Speaking at the opening of a panel discussion devoted to mapping technology at the M:bility | California event, he pointed out just how prevalent mapping technology already is today with low levels of automation.

Continental’s dynamic mapping system eHorizon, for example, has been around for awhile now, alerting drivers to hazards beyond their line of sight. Daimler has harnessed topographical maps as part of the powertrain control input for both cars and trucks. Hyundai and Kia, meanwhile, use information about the grade of the road to help their hybrid  vehicles manage when they use electric power and when they use the internal combustion engine. General Motors is using HD maps from Ushr in its Super Cruise Level 2 automated driving system. Map data in this case is used for geofencing the system and helping it understand what type of speed is appropriate for going into curves in the road up ahead.

Almost everyone in this business is using maps to varying degrees as part of the control system for their automated driving systems, with a few exceptions

Advancing towards autonomy

Maps should prove even more important as the level of autonomy rises. While sensors like camera and LiDAR will help alert AVs to all the visible obstacles in their path, their reach only extends so far. Pivotally, live HD maps—those updated in near real-time—can essentially warn vehicles of what lies out of sight. “Maps not only help you model the world but they serve as that long range sensor,” commented Praveen Chandrasekar, Regional Senior Product Manager, Autonomous Driving, TomTom.

TomTom maps are currently in use by a number of automakers for powertrain optimisation and advanced driver assistance systems (ADAS), but the company is preparing for an autonomous future. “Think about the concept of the horizon. HD maps allow you to look beyond that, beyond the range of the sensors,” added Chandrasekar. “They are also pivotal for operational design domains (ODDs). Most automakers working on automation are setting up their ODDs first on the principle of maps, looking at where they have good mapping coverage and rich attributes.”

TomTom is developing HD Maps

They also add a very important additional level of redundancy. “When you think of automotive, you must always think of safety,” stated T.R. Ramachandran, Executive Vice President of Marketing at Cepton Technologies, a supplier of 3D LiDAR solutions for the automotive, industrial and mapping markets. “There will always be a number of real-life, daily situations where you will need as many tools as possible for as much redundancy as possible. One sensing modality may suddenly not work or may not be optimal for a particular situation. HD maps will be essential for safe driving. I just don’t see the ability to build a truly autonomous system without them.”


Not everyone agrees that HD maps are required for automated driving. Tesla Chief Executive Elon Musk has essentially dismissed HD maps as a crutch, suggesting that street level maps will do all that’s required for AVs. “Most people disagree with that assessment,” asserted Abuelsamid. “At a high level it is theoretically possible but it is not necessarily a path to a robust system.”

He points to a recent white hack as an example. A few months ago, an Israeli security specialist managed to use GPS spoofing to guide an Autopilot-equipped vehicle off its path, convincing it to pull off the road at a different location from where it thought it should go. “If this vehicle had had a supplementary localisation scheme beyond GPS and street level maps, it would have realised that it was not where the GPS thought it was and would not have followed those directions. HD maps as part of an overall robust AV system are probably essential.”

HD maps will be essential for safe driving. I just don’t see the ability to build a truly autonomous system without them

The most common approach in the industry is to double down on redundancy and fuse input from cameras, radar and HD maps. OEMs tend to place more confidence on one data point or the other,” noted Chandrasekar. “Ultimately that boils down to your fundamental view of localisation and path planning.”

Sean Zhu, Co-Founder of PixMoving, backed this up. “Tackling the problem from multiple dimensions creates a safer environment,” he asserted. PixMoving is a start-up developing a Level 4 autonomous vehicle consisting of a multi-functional pod and a self-driving chassis. It has been designed as a shared-mobility solution for use in industrial parks, airports, shopping malls, residential communities, tourist attractions, etc. “Just as you want to have multiple sensing modalities for the core perception,  with some overlap, it is the same for localisation,” he summed up. In PixMoving’s case, that’s an HD Map, sensors and WiFi.

Even for a sensor provide like Cepton, mapping is at the heart of its autonomous blueprint. “Much of our focus is on understanding the corner cases and how we can develop our sensors to improve coverage so they can build higher quality maps,” said Ramachandran. “We take it as a given that mapping will be critical for any autonomous driving application.”

Welcome back , to continue browsing the site, please click here