When industry R&D executives say that autonomous vehicles are going to require a lot more work, they aren’t kidding.
Exactly how much research must go into testing various technologies to ensure that they can be trusted with human lives? How about 8.8 billion miles of test driving.
To surmount that research challenge, some European companies have banded together to address the research as a group, rather than relying on individual corporations and balance sheets to figure it out separately.
Their mission: proposing a guide to vehicle architecture that can handle the industry’s autonomous technology goal.
“We want to reach a global safety reference, that is key,” said Ricky Hudi, chairman of The Autonomous, an initiative that has brought together multiple players in the autonomous vehicle ecosystem. The group includes automakers, Tier 1 suppliers, chipmakers and software developers.
“It’s not that other factors are less important,” Hudi said. “But if you design a house or car or complex system, the architecture is always the first step. And that’s the reason why we kicked off with that.”
The Autonomous was created in 2019 by TTTech Auto. It has members in working groups that are tasked with developing recommendations for a safe system architecture for AVs.
The groups’ output will consist of technical documents that will include recommendations, best practices, guidelines, reference architectures and specifications.
The partners consist of Audi, chipmakers NXP and Arm, the automotive software company of Volkswagen Group called CARIAD, the Fraunhofer Institute for Experimental Software Engineering in Germany, the British artificial intelligence software firm Five AI and the Swedish Royal Institute of Technology.
It is a collaborative approach to tackling what may be the greatest challenge facing the world auto industry, rather than leaving companies to solve the puzzle on their own.
Collaboration will help “accelerate the learning curve” and “reduce development costs,” Hudi said.
The principle of these global safety references is not written in stone, he said, but are recommendations that AV developers can “put their own flavor” on top of.
“You see how complex it is to bring autonomous vehicles on the road. You have to take care of the safety, avoid and mitigate drift, you have to cover all the costs, and you have to drive regulation and standardization,” said Hudi, who is founder and CEO of Future Mobility Technologies GmbH. He also co-founded TTTech Auto, an Austrian producer of software and hardware platforms for autonomous driving, and was previously Audi’s chief executive engineer for electronics.
“You can imagine that this is one of the biggest challenges the industry has today, and we see clear benefits in joining forces.”
Participants in the initiative believe that, through a coalition, it’s possible to develop best-in-class solutions and reduce the risk of going in the wrong direction.
“If many are joining with their bright minds and experience, the
likelihood of working [in] the right direction is higher, you accelerate the learning curve, and you reduce development costs — this is all attractive,” Hudi said.
Iain Whiteside, principal scientist, director of assurance for Five AI, said the development and rollout of automated driving systems — particularly the Level 4 systems that can take full control of the vehicle — are some of the toughest challenges facing engineers.
“We think a deeply collaborative approach is the way forward, instead of each stakeholder developing its own architecture,” Whiteside said. “By pooling the collective wisdom of the major auto and industry players in the working group, we can achieve safer [Level 4] more efficiently.”
Five AI’s roll in the working group will be to conceptually shape the system architecture and define fault-containment and monitoring systems for the subsystems, he said.
Whiteside said that to achieve the 95 percent confidence rate that autonomous vehicles are safer than human drivers in regard to road fatalities, they must be driven for nearly 8.8 billion miles. To put that into perspective, Earth is about 3 billion miles from Neptune.
“The complexity of an automated driving system, and the high-dimensional … space of the real world in which it must both operate safely and be evidenced to meet its safety goals requires a rigorous testing process,” he said.
One of the main benefits of cross-industry partnerships will be to quicken the development of the backbone of self-driving architecture, he said. Consensus across expert partners throughout the ecosystem will help deliver safe autonomy to consumers.
Peter Liggesmeyer, director of the Fraunhofer Institute for Experimental Software Engineering, said that different players have different strengths and weaknesses, and the significant pressure and competition to launch commercial AV services won’t necessarily serve safety.
“We deem comprehensive collaboration across the different autonomous driving stakeholders an important ingredient for making … safe autonomous driving a reality,” Liggesmeyer said. “This should not only encompass industrial companies, but also research institutions such as Fraunhofer, whose mission is to transfer research results, as well as standardization, qualification and legislation stakeholders.”
Liggesmeyer explained that a system is only safe in a particular context. For an autonomous driving system, that context is typically called the operational design domain. That domain defines the environment in which the self-driving vehicle can operate safely.
“Another important element is the ability of the autonomous system to understand the current driving situation from a safety perspective,” he said. The system “has to be able to anticipate potential accidents and related risks and to weigh different planning options against each other, in a similar way as a human driver would do.”
But future systems will not be limited to the driver’s human senses. They will have inputs from a range of sensors, as well as from other vehicles and from traffic infrastructure, including cloud services.
“The complexity and variety of the situations that need to be understood and the features that need to be perceived are very high for self-driving vehicles,” Liggesmeyer said. Designers and R&D engineers must take into consideration uncertainties of perception and the effect they can have on risk prediction, he said.
The collaborative group is focused on how to do that safely.
“That’s the core and the clear reason why we kicked off the initiative,” Hudi agreed. “You have a lot of aspects around safety, but safety is the key. Once you have a proper design of a safety reference, it’s much easier to convince the regulators and legislators, as opposed to doing it the other way around. That’s why even the biggest companies considered joining the initiative.”
Hudi noted it’s important to present an open, welcoming face that encourages other partners to join.
“It’s definitely open,” he said of the effort. “There is no limitation, and that’s important for the initiative. If a company or academia joined the working group, and they feel like they can contribute to this initiative, then they are highly welcome,” he said.
Hudi said the organization will be actively recruiting members on a global scale and said that will help foster networking efforts and bring a corresponding rise in visibility.
“Over time, when companies see our growth, and when during networking events people see how this is producing results, that creates energy and interest to join, and it becomes self-supporting,” he said. “But in the beginning, you have to spend a lot of effort to get the ball rolling.”
The Autonomous is in discussions with Google-backed AV company Waymo and Pittsburgh-based AV startup Aurora Innovation; both companies will support the organization’s main event next month in Vienna with keynote speakers and panelists.
But Hudi admits that not every global power player has been interested in joining the group.
Hudi said that, beyond the technical tasks the working group is focused on, the organization could also help foster agreement on the issue of crafting a common nomenclature around autonomous driving features. That would eliminate confusion in the industry about what system capabilities and limitations really mean.
“Every company has its own way to advertise the product and to give the product an individual flavor,” Hudi said. “But I think it would make sense to align on the same names, the way we have for airbags, or ABS. We have more or less a common language for those systems, and personally, I think it would also be good to label these systems.”