Autonomous vehicles and the imminent death of Vision Zero

For years, the automotive industry has promised that autonomous vehicles will make our roads and commutes safer. Yet road congestion, accidents and driver distraction continue to increase. In fact, U.S. traffic deaths have hit a 20-year high, with no real solution at hand.

Over the long haul, AI promises to ease our burden, drive our cars and enhance our humanity. Yet, recent reports from the U.S. Department of Transportation (DOT) indicate intelligent driving systems may be causing as many crashes as they prevent. The question is not whether we need better control, but rather where this control should rest and how we regulate it. Does control rest with me, my vehicle, the government — or with the off-board servers of corporations that own and orchestrate the data?

A lack of appropriate regulation has left it up to individual citizens — and sometimes states — to decide for themselves how safe they think autonomous vehicles (AVs) really are. But with leading AV automakers skirting the truth and a lack of industry-wide safety standards to hold companies accountable, we’re left asking, “What is the best path forward?”

Addressing this question requires new policies that focus less on the individual software algorithms and more on the emergent roadway effects. Today, the focus is on software engineering processes associated with the AI brain onboard an individual car. Strangely, there’s no focus on the impact of that brain on the roadway around it. Individual intelligence is not wrong, but we need our policymakers to take a step back and consider the bigger, system-level picture.

Why are intelligent cars not improving our roadways?

The most basic challenge facing the community is that we don’t have an objective way to measure performance. We need measurable safety — a performance-based system that focuses on relative motion. This requires that we improve positioning error from meters to centimeters and enhance timing from seconds to milliseconds. Once we have accurate relative-motion data, we can develop an objective, cross-industry metric for analyzing driving performance. Without it, the industry remains drunk on marketing.

The DOT has been attempting to catalog accidents linked to Advanced Driver Assistance Systems (ADAS) and autonomy since June 2021. Original attempts to measure safety and performance were rooted in the DOT’s connected vehicle vision. An ambitious Vision Zero program asserted that, by connecting each vehicle with the ability to send and receive basic safety messages, congestion and accidents could become a thing of the past.

Making this vision a reality would require a few critical ingredients: accurate positioning so the cars knew where they were, and reliable communication to make sure those messages moved quickly and decisively up and down the roadway.

For 20 years, the DOT believed with a fervor that GPS was the key to a connected-vehicle paradise. The plan started to take shape, and tests in open environments worked well, but the grand plan hit a brick wall once testing began in cities. Urban canyons, for example, cause GPS signals to bounce, producing tens of meters of error.

After testing on Sixth Avenue in New York City, the DOT recognized the need for “alternative positioning” to enable its connected-vehicle vision. A phone can have 10 meters of GPS error, and you’ll still find the Dunkin Donuts, but the DOT’s connected-vehicle vision requires a much tighter margin.

With 10cm positioning, we found that predictive braking and coordinated acceleration helped cars move in harmony. With 10 meters of error, however, the cars on Sixth Avenue thought they were driving through Radio City Music Hall. That meant none of the safety apps worked. The DOT thought a little error would diminish the safety benefit only slightly, but instead, it ruined it completely.

Perhaps it doesn’t make sense to reference satellites 20,000 kilometers away in space to get a sense of cars that are only a few meters away. On the other hand, it’s not enough to see only the car directly in front. We need communication to ensure cars move together, anticipating what’s happening up ahead instead of just reacting to what’s directly in front.

Policy leadership for the future of autonomous vehicles

Despite these failed attempts to improve the safety of self-driving vehicles, the industry continues to move forward and explore new approaches. The desire to limit congestion and prevent accidents might lead us to jump right to a specific solution, but the first step should be to agree on a metric for measuring success.

A new approach to measurable safety focuses on relative motion rather than global positioning. It uses the ultra-wideband (UWB) technology that the U.S. government developed for GPS-denied defense applications. Robots used this technology to navigate in tunnels and bunkers, as well as for landmine detection.

At the start, the precision of UWB seemed like science fiction, but recently, UWB has exploded into cellphones and cars, enabling a whole new approach to tracking relative motion. What started in just a few robots now offers the potential to scale massively, enabling an interconnected framework of peer-to-peer positioning and communication.

UWB is just one of several technologies that could meet the specification for measuring safe motion, but currently, it’s the only one proven to work in cities. What is certain is that we can’t keep pretending that GPS solves a problem that it does not. To fully solve the transportation problem, we need a significant number of vehicles to embody the solution, and that could take some time.

However, to measure the performance of ADAS and AVs, a sample set may serve quite well. Thankfully, we don’t need full adoption to gain the benefits of a measurable safety policy. We just need an objective way to measure the safety and performance of a representative subset of ADAS and self-driving cars.

Even so, it is not likely to happen without firm government policy in the form of a mandated data recorder — not unlike a black box on airplanes.

However we choose to enact measurable safety policy, accurate relative positioning must be the underlying foundation. We need to move the bar from meters to centimeters if the resolution of our policy vision is to come into focus.

On top of this foundation, we need investment in smart infrastructure, including edge computing and reliable communication.

Enabling this new system-level approach requires courage on the part of our government officials. The creation of GPS required courage and prescient investment from the government. Publicly available GPS would not have simply emerged from the benevolent self-organization of industry.

Today, we need a new initiative focused on the next big thing: Roadways with measurable safety so we can apply science rather than marketing to the challenge of improving our roads. After all, we can’t effectively regulate what we can’t measure.

David Bruemmer is chief strategy officer at NextDroid.