Do autonomous vehicles need connectivity?

From Tesla’s Autopilot to robotaxis from Cruise and Waymo, the latest generation of automated vehicles are increasingly ‘standalone’ with connectivity only used for non-safety critical functions – but proponents of CCAM (cooperative, connected and automated mobility) argue that AVs can only reach their full potential if connectivity is enhanced. Industry experts take a look at the ideal roadmap to the future.

Not long into his first trip in a San Francisco robotaxi, Tom Alkim eased back in his seat and reflected on the Cruise vehicle’s reassuring smoothness compared to his Uber ride from the hotel. “Not all taxi drivers are cautious and gentlemanly,” he observes. “Sometimes you’re in a hurry and that’s okay. But when you fear for your life, that’s a different story!”

Alkim works at the intersection of traffic management and automated vehicle (AV) development. Before joining Dutch company MAPtm as a Strategic Advisor, Alkim was policy officer for Connected and Automated Driving at the European Commission Directorate-General for Research and Innovation. Based on personal experience of numerous automated rides, he considers the technology increasingly mature.

“Bringing automated mobility to roads involves four different disciplines,” he says. Enabling technology comes first, followed by a legal framework. “If something works and is allowed, it must be integrated into legacy mobility systems, which is where road operators come in. Finally, we want the public to not just to accept but embrace the technology.”

Road operators must consider how infrastructure will accommodate AVs so mobility systems can reap their projected benefits. To operate safely, an AV must remain within an envelope known as the operational design domain (ODD). Only with real-time awareness of ODD attributes like rainfall, traffic density or changes to physical or digital infrastructure can it know when ODD limits are about to be exceeded.

Roadside information could enhance ODD attribute awareness in AVs otherwise reliant on their own sensors. “An example of competing information is ambient temperature, which vehicles already collect,” says Alkim. “Providing that beyond the range of sensors could increase the vehicle’s horizon and proactive capabilities. Traffic managers also have complementary information which vehicles cannot sense, such as traffic densities.”

Vehicles at SAE Level 3 automation are expected to hand control to a human before their ODD is exceeded, while Level 4 vehicles must come to a safe stop by a minimum-risk manoeuvre – expediencies both potentially problematic to road safety. Receiving advance notice of an impassable thunderstorm promises more time and better options than first sensing it as it hits the vehicle.

“OEMs are also interested in work-zone information,” says Alkim. “Relying on their own sensors makes sense to enable first deployments. But as vehicle numbers increase, we enter the territory of traffic management and collective optimization.”

Cooperative or autonomous?

While Dutch national road operator Rijkswaterstaat expects Level 3 AVs on its roads in the coming years, the more complex challenges of Level 4 integration lie beyond its current roadmap. Rijkswaterstaat favours the European idea of connected and cooperative automated mobility (CCAM) versus AVs working in self-reliant isolation.

“We believe in the benefits to society, but want to mitigate the risks,” says Rijkswaterstaat’s strategic advisor for Traffic Management and Smart Mobility, Henk Schuurman. “We believe connectivity can enhance the safety of vehicles which combine their own sensing with information from road operators, such as real-time data about roadworks collected in our C4Safety project.”

In 2020, KPMG rated Dutch roads second in the world for Level 4 AV readiness after Singapore, while an exploratory Rijkswaterstaat study found 72% of its network capable of supporting Level 4 ODDs. But Schuurman is more cautious: “Our highways are well-designed with high visibility and pavement quality. Only 28% are unsuitable for initial deployment, including tunnels and complex tidal flow-lanes.’

When AVs are deployed at scale the Netherlands expects them to use existing infrastructure, minimizing public investments. “We invest in digital infrastructure, but consider major adjustments to physical infrastructure unnecessary,” says Schuurman.

The Traffic Management for Connected and Automated Driving (TM4CAD) project explored mechanisms for infrastructure to contribute to the ODD-attribute awareness of automated driving systems. Funded by the Conference of European Road Directors (CEDR), TM4CAD was led by Henk Schuurman and his former Rijkswaterstaat colleague, Tom Alkim.

“We developed the concept of Distributed ODD-attribute Value Awareness (DOVA),” says Alkim. “DOVA looks at vehicle capabilities as defined by its technology and sensors, the physical and digital infrastructure, then elements beyond our control, like weather or traffic situations. Combining that with the rules of the road determines whether a vehicle is within its ODD and therefore able to drive.”

DOVA provides a useful framework for a dialogue between road operators and vehicle OEMs involved in the Hi-Drive consortium (this will be explored further in Intertraffic World 2025, due for publication late January 2025) initiated by TM4CAD. “But it remains difficult for OEMs to state exactly what information they can use,” says Alkim.

“Usually, OEMs want to build vehicles which can operate independently and don’t want to rely on external information for liability reasons,” he continues. “Let’s say there are trust issues.” Nevertheless, Schuurman believes vehicles must co-operate with infrastructure for the extended visibility needed at higher speeds and sees OEMs starting to buy into this shared vision.

“Information from additional sources can of course make automated driving systems more robust,” says Hi-Drive project coordinator, Aria Etemad of Volkswagen. “We must discuss solutions for new challenges traffic managers may face, like vehicles performing minimum-risk manoeuvres and potentially blocking the road.” Etemad anticipates a two-way exchange with reciprocal benefits.

“Vehicles which accurately sense the environment can share useful information with traffic managers,” he says. “In Sweden, vehicles of different brands send road condition reports to the road operator, Trafikverket. Last year, Trafikverket applied 30% less salt to roads because of this information.”

The Hi-Drive ODD defragmentation concept identifies adverse weather, slippery roads and dynamic traffic rules as automated driving challenges which traffic management centres or vehicle-to-infrastructure (V2I) communications could help solve. But Etemad concedes that vehicle OEMs harbour reservations about relying on external information.

“Our business-case says that if we develop a system, it should work everywhere,” he says. “If vehicles depend on a work-zone warning system which is only installed in Germany, we can only sell them in Germany.” A viable continental market for infrastructure-reliant vehicles requires investment from road operators across Europe, besides Europe-wide standards and harmonized admission procedures which Hi-Drive seeks to advance.

A question of safety

Then comes the question of liability for vehicles relying on roadside information to remain within a safe operational domain. “Our infrastructure friends perhaps need to understand how safety works on our side,” says Etemad. “We have automotive safety and integrity levels (ASIL). The airbags in your car are Level D, meaning they work in 99.99% of cases.”

Just as airbags are held to higher standards than car radios, the reliability required of roadside information would depend on its safety-relevance. Etemad believes work-zone information on which AVs depend may require redundant communication systems to underwrite its reliability. “Then comes cybersecurity, which means another level of investment,” he adds.

Rijkswaterstaat already makes road and traffic information available to navigation service providers and vehicle OEMs via a national data access point. The European ITS Directive makes open data provision mandatory from 2025, but does not specify the required data-quality. “For automation, data-quality is key,” says Schuurman. “Data to which vehicles react automatically must be of outstanding quality.”

“Saying Maybe there’s an incident ahead is no use to OEMs,” he continues. “Relying on outside information is a very strategic decision and we need to define the quality-levels they need.”  The National Access Point Coordination Organisation for Europe (NAPCORE) addresses this with a five-star data-quality rating system.

“Five-star data may be unreasonable to expect all road authorities to provide, but perhaps three-star data is sufficient for initial applications,” says Schuurman. “If only the Netherlands delivers five-star data, OEMs cannot rely on one country. That’s why we collaborate in Europe: to perform together as road authorities.”

Schuurman recently experienced the BMW-Mercedes Level 3 automated lane-keeping system (ALKS), beyond whose ODD limits a human should resume control. Thrusting someone reading a book into a complex driving scenario is assumed to require at least 10 seconds’ notice.

“The 10 seconds is debatable,” says Schuurman. “If you don’t take over, the procedure is stopping in lane, which may create a dangerous obstruction. We’re pushing for a safer stop in the emergency lane and an automatic alert to the traffic control centre, which isn’t mandatory today.”

Level 4 robotaxis or shuttles with no mechanism for human control must perform a minimum-risk stopping manoeuvre when ODD limits are reached. “Potentially, the vehicle says, Hold on, I can’t handle this situation so I’ll just park here!” says Alkim. “Road managers don’t want that happening on the fast-lane of the highway.”

Etemad notes that minimum-risk manoeuvres are already implemented in Level 2 systems for driving in traffic which ultimately stop if they sense a lack of driver engagement. Optimizing the window for human takeover or minimum-risk stopping returns us to the expanded foresight that roadside information could offer.

“Our sensors see about 300m ahead,” says Etemad. “Beyond that field-of-view, accurate information could help us achieve the 10 seconds, or preferably more, needed to hand control to a driver.” Alkim remembers another San Francisco robotaxi drifting into the path of a fire-truck. “We made a wide turn to let it pass, then became a little stuck,” he recalls.

“We reversed then went forward at an intersection, hampered by other vehicles arriving.” It was a moment underlining the avoidable nuisance self-reliant AVs could cause. “The City of San Francisco could say, We’ve had it with your vehicles running into our fire-trucks!” he says. “One mechanism to prevent that could be mandating the use of roadside information.”

This would shift the terms of a dialogue now in its infancy. “Our two different communities first need to understand each other,” says Etemad. “Then we can find the low-hanging fruit, probably starting with use-cases which are not safety-critical.”

This article is a preview of one that will appear in Intertraffic World 2025