Making a real-world impact: Safe and secure software-defined autonomy

Participants included not only leading voices from the world of the passenger cars and commercial vehicles, but also experts in robotics, off-road mobility, drones, and more – reflecting the growing understanding that autonomous technologies are horizontal enablers across industries.  

The 2025 event hosted over 400 participants in person and online, representing more than 230 companies from all over the world. What drew them was the opportunity to learn, share ideas and debate the key issues for autonomous mobility today:  

• • • How to accelerate progress towards safe and secure autonomy 
    • The role of AI in advancing autonomy – and the limits to its role in safety 
    • Industry collaboration, standardization and regulation as well as the role they can and should play in developing effective approaches to safety and security.

In keynote speeches and panel discussions throughout day two of the event, these topics and more gave delegates an expert and thought-provoking insight into the state of play in autonomous mobility and robotics. The Autonomous Main Event also for the first time gave delegates an opportunity to see how innovators and startups are contributing to the adoption of autonomy, with the new Disruptors’ Track, this time featuring HOLON, Roboauto, VicOne and Zendar.  

A new chapter in the history of The Autonomous 

Since it was founded in 2019 by TTTech Auto, The Autonomous has grown into a leading ecosystem for collaboration on safe autonomous mobility. At this year’s Main Event, the Chairman of The Autonomous Ricky Hudi highlighted how the initiative is now ready for its next stage: broadening its scope from automotive into robotics and other domains where safety, security, and robustness are equally essential.  

“Our mission has always been clear,” Hudi said. “The Autonomous is about shaping the future of safe autonomous mobility. Now, we are expanding that vision: to shape the future of safe, secure, and robust autonomous systems across industries through cross-industry collaboration. We strongly believe this unique positioning — particularly from Europe — can deliver valuable contributions on a global scale.” He described how The Autonomous had matured from an automotive-centric initiative, focused on safe ADAS and AD systems, into a neutral, independent multi-stakeholder platform. “We are convinced that scaling robotics requires the same foundations as scaling autonomous automotive systems,” Hudi emphasized. “That is why we are establishing The Autonomous Organization — a truly open and neutral platform.” 

Together with Susanne Einzinger, Head of The Autonomous, Hudi welcomed representatives of the four founding members — Infineon Technologies, NXP Semiconductors, TTTech Auto, and TTTech Group — to the stage for the official launch. As Lars Reger, CTO and EVP of NXP Semiconductors, put it: “It’s great to take part in firing stage two of The Autonomous rocket.” 

The new problem of hardware in software-defined systems  

The expert discussion of autonomy started with the concept of the software-defined vehicle. Many of the biggest and best-known manufacturers in the automotive industry were shaken by its sudden emergence, while much younger Chinese manufacturers have been quick to adopt software-driven architectures. The 2025 Main Event provided plenty of evidence that the rest of the global industry has now fully embraced the software-defined concept. As Jens Kötz, Head of Development Architecture at Audi, said in a wide-ranging fireside chat: “The software-defined vehicle will allow us to develop faster with less effort, to achieve better safety and security, and to scale functionality across models as well as across a vehicle’s systems.”

Joining the fireside chat, Georg Kopetz, CEO of TTTech, highlighted another benefit of a unified software architecture for autonomous vehicles. ”Autonomous systems have to be connected, and that means that they need to enable software updates, including updates which implement new safety features. A software-defined vehicle means that we can incrementally upgrade the car.”  

Kurt Sievers, CEO of NXP Semiconductors gave an interesting insight into the effect this has on his domain, the semiconductor industry. He said: ”The Holy Grail for autonomous mobility is a functionally perfectly safe system, otherwise we humans will not trust it. The interplay of software and hardware is critical for this – we need the closest possible intimacy between the two to achieve true safety. OEMs do not want to do this tight integration of software and hardware, so it’s important that the semiconductor industry provides this as a solution to its customers in the mobility industry.”  

Safety architectures: new thinking from renowned Working Group  

In line with the mission of The Autonomous, safety was the most prominent topic on the agenda. Panel discussions highlighted how OEMs, academia, and suppliers are practically implementing safety features and technologies in mobility systems. 

But the entire industry looks towards the Working Group Safety and Architecture, an initiative of The Autonomous, for high-level guidance. A landmark moment was the publication of the second edition of the Working Group Safety & Architecture Report, unveiled at the event. The group’s goal is to identify and evaluate high-level architectures suitable for a Level 4 highway pilot system, comparing them against criteria such as availability, reliability, security, scalability, and safety of the intended functionality (SOTIF). 

The new report adds three system-level architectures, expands its analysis to seven international safety standards, and introduces a framework for assessing “sufficient independence” of subsystems, an essential concept for ensuring true redundancy. For the first time, engineers now have a structured way to quantify independence between components, helping determine whether redundancy strategies are robust enough for safety-critical systems. Sascha Drenkelforth of TTTech Auto summarized it simply: “Redundancy is not enough — independence matters. Our methodology now helps measure it.” 

Much progress on safety, but academia and experts warn there is much still to do  

In panel discussions throughout the day, the industry’s achievements on safety were celebrated – but experts also acknowledged that answers to some of the problems remain out of reach. In the first panel on end-to-end architectures, for instance, Martin Hart, Director of Software Development at Mercedes-Benz, talked about his company’s introduction of the industry’s first ‘real Level 3 vehicle’, echoing the Working Group in saying that redundancy “is not just about adding a second component — in sensor systems, for instance, you need a second modality.”  

Stefan Poledna, CEO and CTO of TTTech Auto, explained the crucial importance of redundancy. ”In the Western world, we have one fatal accident for every 1 million miles travelled, which from a testing point of view calls for around 8.8 billion miles of road testing for validation – that’s not a practicable option! So, we need redundancy and sufficient diversity, along the end-to-end chain from sensors to processing, communication, power and actuation.”

Redundancy is a strategy for dealing with a component’s failure, but Omer Keilaf, CEO and Co-Founder of Innoviz Technologies, underlined that at Level 4, a LiDAR sensor must provide “100% availability for the lifetime of the vehicle — unlike software, you cannot upgrade the sensor after it has been assembled.” 

Peter Schaefer, EVP and CSO for Automotive at Infineon, highlighted the role of Ethernet and Time-Sensitive Networking in keeping redundant systems affordable. “If redundancy means too many point-to-point connections, the wiring harness cost will be too high. Ethernet-based networks are the scalable solution.” 

From software-defined to AI-defined autonomy 

The day’s second panel asked whether the industry is now moving from software-defined to AI-defined autonomy. Yuping Wang, Head of Marketing Operations at CARIZON, said that ‘we do not see the AI-defined vehicle replacing the software-defined vehicle, we see it as a high-level evolution, from a rules-based architecture to a data-driven architecture’. He cited as an example the ability of an AI-based safety system to recognize a school bus parked by the roadside, and to give the driver a spoken warning to slow down because of the likelihood of children stepping onto the road.  

This CARIZON example shows how AI might enhance safety, but Ronen Smoly, CEO of PlaxidityX, warned that AI opened up a new vector for cyber-attacks. He said: ”There is a lack of understanding of the bad actors who might take advantage of AI. Cars are part of a highly integrated system, from suppliers to OEMs to dealers, and the easiest way into the network is via the car.”  

If AI-based autonomous systems make mistakes or get hacked, public trust in AI will be undermined. As Lars Reger of NXP said in his keynote speech: ”If the AI in my autonomous car starts driving erratically, my kids will not be going to school in the car on their own any more. Trust is the factor which either enables or disables the market for autonomous vehicles.”  

According to Matthias Traub, President and Managing Director of Vector Informatik, ‘there is of course a view from the product perspective and from a customer perspective, but our developers also have to have trust.’ He called for culture change in the software world – ”there is a need for upskilling and greater accountability of our staff, so everyone has the right mindset when talking about AI,” he said.  

Grant Courville, SVP for Products and Strategy at QNX, highlighted another potential safety problem with AI. ”Interrupt latency, scheduling latency, messaging latency – these are all fundamental to software safety,” he said. ”If my system needs to be deterministic but my processing time is variable, I have a real risk. Fortunately, new technology is helping to mitigate this risk, such as the advanced scheduler in TTTech Auto’s middleware.”  

There is also a question over the way that the safety of AI-defined vehicles can be tested or validated. Jeff Walker, Chief Commercial Officer at SDVerse, highlighted one aspect of this. He said: ”Cross-border data governance is a big mess. The Autonomous can play a big role in helping the authorities to get a grip on this.”  

For the frankest appraisal of the safety of AI-defined vehicles today, delegates had the opportunity to hear from Missy Cummings, a former US fighter jet pilot and now a Professor at George Mason University’s Mason Autonomy and Robotics Center. She showed delegates photographs of road collisions caused by vehicles in autonomous mode and said: ”We have a big problem: self-driving cars really do hallucinate – an issue known as the ‘phantom braking problem’.” 

She also stated that today’s autonomous systems cannot have the intuitive grasp of safety that a human driver does. She said: ”Neural networks do not know anything, they do not understand, they do not imagine – they are not thinking agents. And all the end-to-end learning in the world is not going to make this problem go away.”  

Her conclusion? ”There is no such thing as a self-driving car today – they all need human baby-sitters.”  

Legal and regulatory frameworks for autonomous mobility 

Before fully autonomous systems can come into contact with the general public, the regulatory framework needs to be fully established. This was the topic of the third panel discussion of the day, chaired by Benedikt Wolfers, Founding Partner of PSWP. A global perspective was provided by Richard Damm, President and Chairman of the KBA and of the UNECE standard-setting body. He said: ”All countries have the same objective: safety. We need to find ways to work together, to create a common basis for each country’s regulation. This should not be a matter of competition.”  

But Bolin Zhou, Senior Engineer at the China Auto Standardization Research Institute (CATARC), made clear that while China contributes to UNECE standards, ‘we are trying to move at “China speed”, and that means faster than the UNECE’.  

And while regulation is one side of the safety coin, litigation might be the other. In a masterclass, Phil Koopman, Faculty Emeritus at Carnegie Mellon University and a renowned safety engineering expert, asked the question, ”How do we hold computers accountable?”  

He said that traditional product liability is the wrong tool for the job, because lots of companies get away with making bad products. From a legal point of view, he said, the answer is to apply laws which punish negligence on the part of the product manufacturer.  

The threat of litigation is in fact one of the factors which could slow the introduction of autonomous mobility systems to the market. Martin Kellner, Partner at McKinsey and Co., presented survey results showing optimism about Level 2+ adoption by 2035, but warned that legal uncertainty, regulation, and high development costs remain serious pain points. 

Bosch keynote: scaling autonomy for real-world impact 

Following this, delegates heard from Mathias Pillin, CTO of Bosch Mobility, who delivered one of the event’s most forward-looking keynotes. He reminded the audience: “Bringing autonomous systems to life requires two things: technological expertise and the ability to scale globally.” 

Bosch, he said, combines decades of engineering excellence with leading AI technology to create systems that are safe, reliable, and deployment-ready. But true progress will only come if the industry collaborates on data, tooling, standards, and the explainability of AI — while still allowing differentiation where it creates customer value. Pillin concluded confidently: “If these challenges are addressed, autonomy will reach the market far sooner than many expect.” 

Autonomy beyond automotive 

In line with the expansion of the scope of The Autonomous’ mission beyond vehicles, the day’s final panel debated the rise and evolution of robotics. While much of the technology for safe autonomy is shared with the automotive world, the application of it can be very different when mobility is not tied to the road.  

As Muneyb Minhazuddin, Customer Growth Officer at Ambarella, said: ”As soon as you introduce humans, you get into safety, regulation, and all the rest of it. In a pure machine-to-machine environment, you can design your autonomous system from first principles.” He proposed more segregation of autonomous systems away from any human presence, which he said would allow for much faster development and more innovation.  

The market environment for robotic operation is in many ways more friendly outside the automotive arena. As Christian Poschmann, SVP at agricultural equipment manufacturer CLAAS, said: ”Farmers trust autonomous systems – that’s not the problem.” Psychological research organized by Andreas Kugi, Professor at Vienna University of Technology, has shown that people are happy to replace human-operated forklift trucks with autonomous trucks. But he said: ”They want to be able to see what the machine is doing, so it’s important to provide a video feed when the truck leaves the operator’s direct line of sight.”  

The panel also provided an interesting insight into the different safety requirements of different types of robots. For instance, Yuri Steinbuch, Chief Business Officer at Avular, which makes robotic and drone technology, pointed out that safety in aviation is hard because there is no emergency stop switch for use in the air. He said that robotics gave scope to increase autonomy in increments. ”We are starting with slow machines, such as cleaning robots, and learning from them, so that in time we can implement autonomy at higher speeds and in more crowded environments,” he said.  

Looking ahead 

The 2025 Main Event closed with recognition that the collaboration fostered by the new The Autonomous Organization is helping to advance not only technology but also professional understanding of the technical, commercial, legal and social questions arising from autonomy. And delegates will have the opportunity to learn even more when the seventh edition of The Autonomous Main Event takes place, again in Vienna on 23-24 September 2026.