Safety remains at the top of the agenda for the AV industry

In this fifth anniversary year of The Autonomous initiative, this theme of safety is drawing more support than ever: for the first time, The Autonomous Main Event was extended from one to two full days of workshops and conference and attracted record numbers of participants: more than 580 people attended the event in person or virtually, and more than 250 companies accepted invitations to attend the event.

 

This unprecedented level of participation reflects both the exciting opportunities in autonomous mobility, but also growing understanding of the scale of the challenge facing the industry if it is to gain broad public trust. The Autonomous Main Event drew together partner companies from around the world, such as NXP Semiconductors, Mobileye, Infineon, Thoughtworks, Elektrobit, TTTech Auto, McKinsey & Company, Audi, DENSO, Green Hills Software, HAL4SDV, Federate, Innoviz Technologies, ABA (Austrian Business Agency), TTTech, SDVerse, as well as various academic institutions.

 

Dirk Linzmeier, CEO of TTTech Auto, summed up the state of play when he said: ‘We are in the middle of the biggest transformation in automotive history: the industry is investing heavily in developing different powertrains, in the move to the software-defined vehicle (SDV), and in de-globalization. At the same time, market shares in China decrease and profit declines significantly for Western OEMs. For autonomous driving, safety is the number one concern for consumers. What we know is that partnerships and ecosystems are crucial for success in safety design. Safety is a realm for neither competition nor compromise – it’s about getting it right.’

 

The effort to collaborate is led by The Autonomous, for instance through its Working Group on Safety and Architecture, which reported on its work over the past year. Georg Niedrist, Senior Fellow at TTTech Auto, described the methodical approach to safety architecture design which underpinned the report that the Working Group published in December 2023. The report identifies, describes and evaluates logical architectures, as well as their implementation in both physical hardware and software. The group found that asymmetric architectures which allow for a ‘majority voting’ approach are suitable for autonomous driving systems.

 

And the group’s work continues. Georg Niedrist said, ‘The working group is now developing a second edition based on five follow-up work packages on sufficient independence of sensor systems, security, validation of high-availability infrastructure, post-deployment observation, and additional conceptual system architectures.’ It is hoped this second edition will be ready for publication at The Autonomous Main Event in September 2025.

 

Further insights into architectural choices came from a panel discussion on ‘Safe paths: building safe and fail-operational architectures for autonomous vehicles’. The discussion was moderated by Phil Koopman, Associate Professor at Carnegie Mellon University. His challenge to the panel: ‘We know how to build systems that shut down safely on failure. But turning the car off in the middle of the road does not work.’

 

Andrea Bondavalli, Professor of Computer Science at the University of Florence, admitted that ‘we are not equipped for this in the car today. Standards such as ISO 26262 deal with safety, but they do not show how to continue operating in the presence of a failed system. I think that we could learn from the aerospace industry in this regard.’

 

The panel stressed the importance of understanding all the potential causes of failure. As Mathias Pillin, CTO of Bosch Mobility, said, failure can occur not only in the car’s brain or communications network – ‘watch out for the power system as well,’ he said. ‘The cable harness can be a source of failure, so we need a redundant way to route power around the vehicle.’

And even redundancy is not enough. Professor Bondavalli said: ‘In addition to redundancy, you need diversity, so that the redundant component does not simply replicate the error which disabled the primary component.’

Pioneering Safety Research: Insights from Mobileye and Waymo

Behind the scenes, intensive research efforts are deepening the industry’s understanding of the safety challenges. Shai Shalev-Shwartz, CTO of Mobileye, presented the Main Event with a meticulous description of his company’s approach to safety in mass-market autonomy. He proposed a Responsibility-Sensitive Safety model to tackle the ethical tension between safety and usefulness, addressing questions such as, what is the maximum reasonable speed in built-up areas? He also discussed how to make the principle of redundancy applicable to AVs. He said, ‘The familiar “majority voting principle” is not always applicable when arbitrating between different sensor systems in an AV. Instead, we propose a Primary-Guardian-Fallback model. Here, if the guardian dictates the primary input is valid, the valid is chosen. Otherwise, the system chooses the fallback input. This results in a very high mean time before failure (MTBF).’

 

Like Mobileye, Waymo is pursuing deep research into AV safety. Trent Victor, its Director of Safety Research and Best Practices, urged the whole of the industry to invest in research and – crucially – to follow Waymo’s lead by adopting an ‘open science’ approach. He said: ‘Safety needs appropriate communication so that ordinary people understand it. We need a better discussion around safety. That’s why Waymo publishes scientific papers about safety, and papers on best practices such as for injury risk estimation – the models that the industry needs for this do not exist today.’

Navigating the Regulatory Landscape

Questions about society’s tolerance of risk in the operation of AVs will in practice be largely decided by governments, through legislation and regulation. A panel discussion on ‘Navigating the regulatory landscape’ addressed the need for global harmonization of regulation governing advanced autonomous vehicles. Richard Damm, President of the Federal Motor Transport Authority (Kraftfahrt-Bundesamt, KBA) in Germany and Chairman of the UNECE Working Party on Automated/Autonomous and Connected Vehicles in Geneva, told delegates that the UN was aiming at the release of regulation on Level 4 autonomy ‘in mid-2026’.

 

Before then, individual manufacturers are forging their own path towards AV development. Stefan Oexl, VP of Engineering at Torc Robotics, is involved in the testing of autonomous heavy trucks. He said, ‘Our preferred place for testing is where we want to first enter the market, and that needs to be in a friendly environment from a regulatory point of view. For us, that includes Texas, on the highway between Dallas-Fort Worth and the Mexican border: there’s little rain, no snow, and the traffic is much less complex and more predictable than elsewhere.’

The panel’s consensus? That the application of Level 4 autonomy is much more likely in commercial vehicles in the short term than in private vehicles. Nevertheless, as Benedikt Wolfers, panel moderator and Founding Partner at PSWP, noted, ‘regulation is not a hindrance in the European Union for the AV space; it will push technology innovation.’

Autonomy in the era of the software-defined vehicle

The Autonomous Main Event took care to place the discussion about safety in the context of other big trends driving the AV industry, of which perhaps the most important is the software-defined vehicle (SDV).

 

Prashant Gulati, CEO of SDVerse, confirmed this with his remark that ‘90% of cars built in 2030 will be SDVs’. He added: ‘Every car has between 25 and 100 ECUs, so we need software to be decoupled from this complex hardware. The heavy lifting in the next decade will be the architecture change from domain-based to zonal or centralized architectures. He quoted Warren Buffet who said that the most important job of any CEO is the most effective allocation of resources – insourcing vs. outsourcing, identifying what is core differentiating and what not, understanding that over time a lot of SW can become commodity, and thus encouraging reuse. The move to software-centric development is inevitably going to lead to culture change in the automotive industry – not that this will be easy. Michael Fait, Head of SDVs (Europe) at Thoughtworks, pointed out a fundamental difference between software and hardware developers: ‘Software developers think, “The software is never done.” But car companies’ approach is “Right first time”. It can be easy to oversimplify, and to say “We just need to change the culture.” But this means addressing real-world metrics, and forging a common view on issues such as time to market.’

 

Stefan Poledna, CTO of TTTech Auto, stressed the importance of hardware and software development teams working in tandem on SDVs. He said: ‘Safety needs to be implemented through the whole stack. And this needs a proper architecture in place. Safety cannot be built up incrementally – that approach creates big problems in terms of time and cost. At TTTech Auto we prefer to talk not about SDVs but the 4SDV: systems, security, safety and software.’

 

In his keynote, Lars Reger, CTO and EVP of NXP Semiconductors, extended the discussion beyond just automotive hardware and software, describing the concept of ‘the great robot awakening’.

He pointed out that the challenge for automakers was to navigate the mounting complexity of software and hardware integration. He said: ‘ECUs contain from tens to hundreds of hardware and software components. Features and variants of car models drive further complexity. What the industry needs instead is a single hardware platform to decomplexify and ease integration.’ He concluded his speech by saying that we often overestimate the next 2 years and underestimate the next 10 years.

Driving Future Mobility – From the Road to the Cloud

In his keynote, Nikolai Setzer, CEO of Continental, pointed out that the SDV offers a safe, exciting, and autonomous mobility experience by controlling, monitoring, and maintaining vehicle functions through software. Autonomous driving will benefit from innovative vehicle data usage, cross-domain software architecture, centralized hardware, and secure cloud connections. Partnerships are crucial for achieving autonomous mobility safely and efficiently.

Collaboration generating answers to safety questions

Fortunately, The Autonomous Main Event provided space for a diverse range of players to collaborate on approaches to this and other problems around SDVs and AVs. For instance, the Day 1 workshop on ‘Safety of embedded AI’ developed a consensus, according to Juergen Schaefer of Infineon, that AI is playing a crucial role in planning and control functions in autonomous driving. He said: ‘By extending classical planning and control functions through AI, we can gain a 50% improvement in tracking accuracy, a 10% improvement in energy efficiency, and 100x lower compute usage.’

 

Elektrobit hosted a second workshop on ‘Enabling SDV: open source from cloud to cockpit’. Wolfgang Thieme of Elektrobit reported back on participants’ views on the use of open-source software in SDVs. He said: ‘Participants strongly agreed that open-source software will be a key component of SDVs in future, but that providing warranties and support for open-source products is a huge challenge. They also strongly agreed that cloud technologies will be a key contributor to SDVs in future.’

 

The third workshop, hosted by McKinsey and Company, was on ‘The way to higher profitability with ADAS’. It found that ADAS functionality is a key factor affecting buyers’ choice of car. But Martin Kellner, Partner at McKinsey, said: ‘The pain point for the automotive industry in ADAS is the combination of high cost and low profitability. The workshop found that the high costs are mostly attributable to integration, test and validation, and the high cost of software development. Most participants believe that a solution is to be found in the separation of hardware and software, to isolate the system-on-chip hardware, middleware, and application software.’

 

Furthermore, the insightful fireside chat between Aurora’s Co-Founder & Chief Product Officer, Sterling Anderson and Nils Jaeger, President of Volvo Autonomous Solutions highlighted how these two companies’ combined safety expertise will pave the way for a new mode of transportation: safe and autonomous trucks designed to support societal growth and save lives. According to Anderson, ‘the Aurora Driver is engineered not only to perceive the world but also predict it. It sees three times further than humans and twice as far as conventional lidar. This system is always on, never drowsy, distracted, or drunk, constantly observing its environment to avoid accidents that humans might not.’,

 

Georg Kopetz, CEO and Co-Founder of TTTech is also convinced that the era of robotics will help address societal challenges. While moderating the panel ‘Autonomy beyond Automotive’, he mentioned: ‘With the average Japanese farmer being 68, implementing assistance systems or full automation is crucial. Together, we can tackle significant societal challenges through technology and innovative business models. I am excited about building global bridges across continents and industries.’

 

One industry, one goal – and no corners cut

In every effort to devise architectures for safe autonomous driving, to steer regulation or to develop safe road infrastructure, the automotive industry will move faster if it moves as one. And The Autonomous Main Event provides the industry with a forum for exactly this kind of collaboration.

 

Moreover, The Autonomous initiative thrives on its Working Groups and Expert Circles year-round, making it more than just a conference. It actively engages the industry’s players to collaborate on AV and SDV challenges, with their findings showcased at the annual flagship event. As the Chairman of The Autonomous, Ricky Hudi, pointed out, ‘More than ever, this goal of safety will determine the success of an automotive company. And it cannot be done alone – it requires collaboration.’

 

Joint working was much in evidence across the panels and workshops, and in the contributions from the Main Event’s array of distinguished experts. The Autonomous initiative is set to expand, with plans to establish more working groups on various topics and grow its global ecosystem in 2025. The industry eagerly anticipates further collaboration opportunities when The Autonomous Main Event reconvenes on September 17-18, 2025, where the latest findings will be presented.