Keysight Technologies and the Centre for Assuring Autonomy (CfAA) at the University of York (UK) have announced a research collaboration to advance the safe deployment of artificial intelligence (AI) in software-defined vehicles (SDVs). The collaboration will focus on developing practical methods to validate AI systems and generate the evidence needed to demonstrate their safety and reliability. The work is designed to enable auto makers and suppliers to meet emerging industry requirements while reducing development risk and accelerating the deployment of AI-enabled automotive systems.

Keysight and the CfAA say that together they will help bridge the gap between AI safety research and practical engineering by advancing methodologies for structured AI safety cases and building justified confidence in the safety of AI-enabled automotive systems.

The research will focus on creating evidence-driven approaches to support the implementation of ISO/PAS 8800 requirements, exploring measurable safety-scoring methodologies grounded in academic research and industry standards, and designing practical frameworks for generating auditable AI safety evidence. The resulting methodologies are intended to support the process of building AI safety cases, improve confidence in AI validation, and help engineering teams demonstrate justified AI safety more efficiently throughout the product lifecycle.

By combining the university’s academic expertise with Keysight’s AI validation techniques, the collaboration seeks to advance methodologies applicable to automotive AI development.

Simon Burton, chair in systems safety at the University of York, said, “The automotive industry is at a pivotal point where AI technologies are becoming increasingly integral to vehicle functionality. Ensuring these systems can be evaluated using robust, evidence-based approaches is essential. This is where the CfAA is ideally placed to support Keysight. We have produced several freely accessible frameworks and guidance already being used by industry safety professionals in the transport sector. This collaboration is another way we are supporting the advancement of practice methods that help translate AI safety principles into engineering practices that can be applied in safety-critical environments.”

Lukas Klose, head of the automotive AI Solution Center at Keysight, added, “Automotive organizations need practical and scalable ways to build confidence in AI-enabled systems. By combining leading research in safety assurance with Keysight’s holistic AI Validation Framework, we aim to develop methodologies that help engineering teams generate structured evidence for AI safety cases and support the deployment of trustworthy AI technologies in conformance with international standards such as ISO/PAS 8800.”

The research is expected to inform future development of Keysight’s AI Software Integrity Builder, strengthening support for AI safety arguments, evidence generation and validation aligned with automotive industry expectations.

In related news, SeoulTech develops AI framework to assess vision-system robustness

In computer vision and robotics, ensuring that AI systems remain reliable under real-world conditions is a growing challenge. Deep neural network-based vision systems are increasingly used in safety-critical applications such as autonomous driving, where misinterpreting a traffic sign could potentially lead to unsafe decisions. Everyday wear and tear can subtly alter traffic signs, raising questions about whether naturally occurring damage could also expose vulnerabilities in AI-based recognition systems.

To understand and address the problem, a research team led by associate professor Seong Tae Kim from Kyung Hee University and assistant professor Hong Joo Lee from SeoulTech developed Adversarial Wear and Tear (AdvWT), a framework that exploits natural wear and tear as an adversarial signal.

“We focused on traffic signs because they are exposed to weather and environmental damage throughout their lifetime, and their accurate recognition is essential for safety-critical applications,” explained Dr Lee. “Unlike temporary optical attacks, natural deterioration can persist until a physical object is repaired or replaced.”

To create AdvWT, the team trained a generative image-to-image translation model to learn the visual characteristics of damaged and undamaged traffic signs. The model, based on StarGAN-v2, learned a latent “damage style” representation that can reproduce diverse forms of realistic deterioration while preserving the identity and meaning of the sign.

By progressively adjusting the damage representation, they generated signs that looked naturally degraded but were more likely to be misclassified by an AI system. In a human study involving 32 participants, AdvWT-generated images received high naturalness ratings, closely matching the perceived realism of real damaged traffic signs.

The framework was evaluated against two traffic-sign datasets and tested against eight recognition architectures. Across the evaluated models and datasets, AdvWT achieved near-perfect attack success rates on lightweight CNNs such as ResNet-18 and MobileNet, while also remaining effective against transformer-based models. AdvWT also achieved the highest average transferability across most tested model combinations, suggesting that the adversarial perturbations could transfer across different model architectures.

To test the framework in a physical setting, the researchers printed clean and adversarial speed-limit signs and photographed them under different distances, viewing angles as well as indoor and outdoor conditions. The resulting images remained effective at misleading the traffic-sign classifier, showing that the adversarial effect persisted after printing and recapturing under varied physical conditions.

The researchers also demonstrated that the same bidirectional model could be used to restore naturally damaged traffic signs, suggesting potential applications beyond adversarial testing. Importantly, training models with AdvWT-generated damaged signs improved their ability to generalize to real-world damaged traffic signs, suggesting that simulated natural deterioration could be used to identify and strengthen weaknesses in vision systems.

“Building reliable AI requires more than improving average performance. It requires continuously identifying where AI systems fail, understanding why those failures occur, and using those insights to make the systems more robust,” said said Dr Kim. “Over the next five to 10 years, research in this direction could play an important role in developing AI systems that can be deployed more reliably in real-world, high-stakes domains such as healthcare and finance.”

Recent news, AI co-driver turns spoken observations into structured test data

The IATF 16949 certification covers E-Con Systems’ automotive camera portfolio across four camera types: forward-facing cameras, rear-view cameras, surround-view cameras and driver monitoring cameras, spanning the core vision systems used in mobility applications. E-Con Systems’ automotive cameras are also validated on leading platforms like Nvidia Jetson, supporting a proven path from prototyping through production.

IATF 16949 certification reflects improvements across E-Con Systems’ quality management processes, including risk-based thinking throughout the product lifecycle, product and manufacturing risk mitigation, and structured controls for safety-critical characteristics. The certification also covers customer-specific requirements, error prevention, design-for-manufacturability, APQP and PPAP processes and enhanced quality management across the supply chain.

“As autonomous vehicles and intelligent machines increasingly depend on vision to perceive and respond to the world around them, the quality and reliability of every camera we design and manufacture become mission-critical,” said Ashok Babu, president at e-con Systems.

“The IATF 16949 certification reflects years of discipline built into how we design, develop, and manufacture our vision systems – giving automotive customers a design-responsible partner who identifies and addresses risk early in the product lifecycle, not after it reaches the field.

“We are now extending that rigor from quality to safety: we are actively developing ISO 26262 functional-safety-compliant cameras for automotive and mobility applications. Together, IATF 16949 and ISO 26262 reflect a single commitment – automotive-grade quality in how we build, and functional safety engineered in by design.”

Recent news, AI helps Hyundai unlock decades of crash test data

Backed by UK government funding through the CAM Pathfinder program, this new technology combines Gaussian scene reconstruction and generative AI techniques to create highly realistic digital representations of real-world environments. The company has been creating and using simulation software for automotive manufacturers since 2015.

Gaussian scene reconstruction converts images captured from the real world into photorealistic 3D environments that can be rendered efficiently in a simulation. It allows developers to recreate roads, junctions and other driving environments with a level of realism that is otherwise difficult to achieve at scale.

The project will also apply diffusion-based AI models to modify and enrich these digital environments. This enables the rapid generation of variations in lighting, weather and other scene conditions, allowing autonomous vehicle systems to be tested across a much wider range of scenarios than could be collected manually.

Many cars on the road today use Level 2 driver assistance, where the car can control the steering and speed but the driver must remain alert at all times. Level 3 is also available in some cars. This new technology will be used to develop technology for cars that will use Level 2++ and Level 3 automation. Five’s simulation tool will work with others within Bosch Mobility to create a comprehensive platform. In addition to Gaussian scene reconstruction and diffusion, it will use state-of-the-art artificial intelligence to reconstruct photorealistic digital twins of the world. It will be possible to create scenarios and evaluate how the car handles them. Ultimately, the simulation will be able to train and validate L4 autonomous vehicle systems.

Dave Buffey, head of Bosch Mobility UK, said, “This testing software will be an important cutting-edge innovation that will be developed by Five AI’s team. Our strategy is to work closely with Five to solve the complex challenges that our car manufacturer partners are tackling as they incorporate the latest autonomous driving tech into their vehicles. This is one of the biggest challenges in today’s automotive industry.”

In related news, Arnold NextG aligns Safety-by-Wire with UNECE ADS rules

The UN regulation establishes an approval framework under the 1958 Convention; the UN GTR creates a globally harmonized technical basis under the 1998 Convention. This does not result in a globally identical set of registration laws, but rather in a significantly more common technical standard for safety, validation and verification.

“Autonomy is not resilient simply because an algorithm makes the right decision,” said Kevin Arnold, CEO of Arnold NextG. “It is resilient when the vehicle executes that decision in a controlled manner even under relevant fault conditions. The new ADS regulations do not prescribe a specific architecture. However, they make it clear that safety must be demonstrated across the entire chain of effects and the entire lifecycle. That is exactly where our Safety-by-Wire technology comes into play.”

The new ADS regulations clarify that the performance of an ADS alone is not enough. Only the safe electronic control of steering, braking and propulsion creates the conditions necessary for autonomous driving decisions to be reliably translated into vehicle movements. According to NextG, Drive-by-Wire thus becomes a central technological foundation for fully automated vehicles.

The safety chain does not end with the AD stack

Digital driving decisions must be translated into steering, braking and propulsion commands, validated, prioritized, monitored and, in the event of a fault, safely degraded.

Arnold NextG describes this safety-critical level as the Control Layer. The term is not a regulatory requirement but rather an architectural classification.

The Control Layer connects the decision-making logic with the actuators and creates a defined, verifiable interface between software and physical systems. It must function independently of whether a driving command originates from an ADS, a teleoperation control center or driver interaction.

Safety-by-Wire as the Control Layer

With NX NextMotion, Arnold NextG is developing a modular, vehicle-platform-independent and multi-redundant control platform for drive-by-wire. The quadruply redundant hardware and software architecture is designed to control primary functions such as steering, braking and propulsion, as well as selected safety-critical secondary functions, on a common system basis and to maintain operation in the event of defined single faults.

According to the company and the TÜV certifications provided, the safety concept is aligned with functional safety standards ISO 26262 up to ASIL D and IEC 61508 SIL 3. Arnold NextG also provides TÜV certifications for steering and braking systems in accordance with UN ECE R79 and R13, as well as road approval for NX NextMotion autonomous, teleoperated and software-defined vehicle projects.

Regulatory requirements become part of the operating model

At the 21st session of the UNECE Working Group on Automated Driving Systems (ADS), work continued on the Guidance and Interpretation Documents (GID) for UN R185 and UN GTR No. 26. The session emphasized that the new ADS regulations are accompanied by technical guidelines and interpretive aids and that their practical application is being gradually refined.

From Arnold NextG’s perspective, this development underscores that requirements for the safe implementation, monitoring and verification of automated driving functions must be considered across the entire vehicle lifecycle. It reinforces the company’s view that regulatory requirements should be considered from the outset of vehicle development rather than addressed after technical work is complete. This includes integrating requirements for verification, diagnostics, data management, software updates and operational monitoring into the system architecture and the product lifecycle.

In related news, Curiosity Lab launches program to support growth-stage AV companies

The system combines an FPGA core for real-time data processing with interchangeable SerDes (serializer/deserializer) interface devices, enabling engineers to reconfigure test setups as display and camera requirements change.

The launch is the latest addition to Emerson’s NI PXI platform, an open, modular test system, and extends its capabilities for automotive testing applications. The system includes the NI PXIe-1480 FlexRIO multiprotocol vision carrier module, which supports interchangeable interface devices for testing automotive displays and cameras. The first available interface devices support FPD-Link IV, a high-speed automotive video interface used in ADAS and in-vehicle display applications.

As vehicle architectures shift toward centralized electronic control units, engineers must validate multiple high-speed interfaces on increasingly complex systems. Conventional display test methods, often relying on cameras or visual inspection, limit scalability and introduce variability. By providing a direct interface to ECU and display SerDes input/output (I/O), the modular PXI-based approach improves test coverage through automated, deterministic testing. The result is a single, flexible platform that streamlines validation and improves test repeatability.

“The increasing complexity of automotive electronics requires a more flexible and scalable approach to validation,” said Sam Burhans, chief product manager with Emerson’s test and measurement business. “This modular architecture allows engineers to adapt their test systems to evolving interface standards without redesigning their entire setup.”

The NI PXIe-1480 carrier module serves as the foundation for this architecture, supporting interchangeable interface devices tailored to specific testing requirements.. Emerson plans to extend the platform later this year with additional interface devices supporting a broader range of camera and display protocols.

The solution is designed for engineers in automotive validation and production test environments, including Tier 1 suppliers and original equipment manufacturers, supporting applications across in-vehicle infotainment, ADAS and autonomous driving systems.

Recent news, Beamr partners with Intempora to bring ML-safe video data to RTMaps

Beamr Imaging and Intempora, a dSPACE company, have announced that Beamr’s ML-safe video data technology is now available on the RTMaps AI Store, ready to use across the AV stack. The collaboration lets AV teams reduce their data volumes while preserving ML accuracy, without changing the stack they already rely on.

Real-Time Multisensor Applications (RTMaps) is a modular development and execution middleware, designed to acquire, synchronize, process, record and replay heterogeneous data streams in real time. Beamr’s ML-safe compression can be deployed at the stages where video data accumulates: at logging, where compression lets constrained storage hold more recorded footage; in the cloud and data center, where it cuts existing petabyte-scale storage and transfer costs; and in simulation and training, where it reduces the real-world and synthetic datasets feeding hardware-in-the-loop (HIL) testing.

At each stage, Beamr’s content-adaptive bitrate technology (CABR) reduces file sizes by up to 50% while the object boundaries, edges and scene detail that perception and detection models rely on are preserved. CABR reduces file sizes beyond standard encoding alone, and runs GPU-accelerated for high-throughput processing.

“The RTMaps AI Store gives AV teams ready-to-use software components they can easily drop straight into their development pipelines, and compression is a capability our users have been asking for,” said Nicolas Du Lac, CEO of Intempora. “Beamr is the first compression technology in the store. It lets teams significantly cut the size of their video data without compromising the models that data feeds, inside the framework they already work in.”

“Managing the volume of data AV programs generate is a constant challenge for the teams we work with, and video is the largest part of it,” said Jacob Perrin, ADAS/AD engineering manager, dSPACE. “In our testing, Beamr reduced that video data while preserving the detections perception models depend on – which is what makes it the right fit for the petabyte-scale workflows across our ecosystem.”

“AV programs run many video pipelines, real-world capture, simulation, synthetic data – each feeding different models – and all need to meet their accuracy targets,” said Sharon Carmel, Beamr CEO. “Our content-adaptive compression achieves improved results over existing workflows, so the same technology reduces data volumes through the AV lifecycle. On the RTMaps AI Store, Beamr’s technology fits into existing pipelines without changing how downstream models behave. AV teams want certainty before making changes, and our Beamr Blueprint methodology provides them an end-to-end assessment of their own video workflows.”

In recent news, Yanfeng reveals XiM27 smart cabin concept with reconfigurable layout and AI-adaptive seating

Through the program, selected companies may receive funding ranging from US$25,000 to US$250,000 to support deployments on Curiosity Lab’s autonomous mobility route in Peachtree Corners in the US state of Georgia. The program is designed for companies with technology that is ready to move beyond the lab, beyond the closed course, and into live operating conditions.

Curiosity Lab is a living laboratory for place-based deployment in metro Atlanta, built to help companies test next-generation technologies in an active city environment. The autonomous mobility route includes a nearly four-mile smart mobility loop along Technology Parkway and State Route 141, with dedicated autonomous vehicle lane segments, shared and mixed-traffic roadway, enhanced sidewalks, rights-of-way, smart city infrastructure and real destinations throughout the corridor.

The test lab recreates a real city environment with active traffic, intersections, sidewalks, business entrances, commercial activity and public infrastructure. For autonomous mobility companies, it creates an opportunity to validate technology in the kinds of conditions that matter for commercialization, customer adoption and scale.

Curiosity Lab is looking for growth-stage companies and innovation teams working on autonomous mobility technologies, including autonomous passenger shuttles, low- or high-speed autonomous vehicles, autonomous delivery vehicles, sidewalk or curbside delivery robots, autonomous logistics and last-mile mobility systems and robotic mobility systems.

Applications must be submitted by 11:59pm Eastern Time on July 24, 2026. Selected projects will take place in Peachtree Corners between October 1, 2026, and September 30, 2027.

Related news, Hydrogen-powered JCB Hydromax completes UK testing

To this end, the company has joined the Software Defined Vehicle (SDV) Working Group under the Eclipse Foundation and will participate in the S-Core project to develop an SDV software platform in earnest.

The S-Core project is a global initiative launched primarily by European companies in late 2024 to standardize foundational technologies such as software platforms and middleware. It is the first open-source-based software platform development project to meet ASIL-B, the automotive industry’s functional safety standard.

Currently, a total of 13 companies are participating in the project and prioritizing the implementation of core technologies required for SDVs, as standardized foundational technologies are essential for accelerating the development of applications such as autonomous driving. Under the common objective, participating companies are also working to prevent redundant investments while improving system stability.

The most distinctive feature of the S-Core project is that it applies an open-source development approach, previously used primarily in the IT industry, to the mobility sector. Participating companies disclose some of their software technologies, enabling developers around the world to freely use and improve them.

Another reason companies disclose coding technologies that constitute intellectual property is the potential to create both tangible and intangible added value. By encouraging more developers to use their software, companies can increase the likelihood of their technologies becoming global standards.

The technology Hyundai Mobis plans to disclose is a so-called ‘container solution’ that minimizes interference between software within the Linux operating system. The technology effectively creates partitions between a wide range of software in SDVs, packaging them individually so they can operate quickly without affecting one another.

The container solution is known to be more than 10 times faster than existing technologies in automotive controller environments. The company has also secured an always-on integrity assurance function to prevent software tampering caused by external intrusions and other risks.

Related news, Next-generation Helm.ai models deliver full-HD 360° synthetic driving environments

“It was a great start yesterday (May 19) with the launch of our smart prototyping concept with lots of positive feedback from customers and partners,” said Tanneke Reinders, executive VP at HBK. “From this morning’s presentations we really see ‘smart testing’ coming alive. The key is transforming from data creators to data interpreters – that’s the main shift for us. It’s about connecting data in one ecosystem. When we talk about data, especially the amount of data and fragmentation of it, we want to support.”

This year there were multiple tire-focused presentations, highlighting the critical role of the tire and its relationship with the vehicle. It seems tire and vehicle developers are working closer together than ever to refine their use of simulation and tighten the integration between the digital and real worlds. This was exemplified in the opening presentation by Alessandro Capobianco, R&D specialist for tire models and vehicle/DIL simulation at Bridgestone, and Eric Walter, vehicle dynamics CAE engineer at VW (both above), who provided an insight into their vehicle/tire co-development work.

Another of ATTI’s highlights from the summit was the Nissan and Horiba Mira presentation (right), in which the two speakers, Leo May, lead simulation engineer, Nissan Motor Corp, and Stratos Stratoudakis, a consultant at Horiba Mira, detailed how they are working together to prioritize virtual development by moving it further forward in the cycle, with the ultimate aim of reducing development time by 25%.

Marc Hedrich, a development engineer at event exhibitor Team Rosberg Engineering, said of the company’s attendance at the summit: “We are presenting our expertise in providing simulation models. The focus for us is to learn new things because software is ever evolving and you don’t want to be left behind. I like this year’s approach of trying to get the best data and not the most. I’ve learned already how others such as Mercedes-AMG are using these tools. It’s cool to connect with companies that are doing similar things and others that are doing things we’ve never heard of.”

This morning concluded with a presentation from the rarely heard-from Tesla (top), which gave insights into the company’s NVH analysis using a simulator.

Francesco Calabrese, who was attending as an exhibitor representing Fraunhofer ITWM as well as the institute’s spin-off, Virtual Tire Technologies, said, “It’s a good event to push the word for using more simulation with the goal of spending less money on prototypes. What I like is the excitement in this field as it grows – every year there are significantly more people which is a sign people are understanding the importance of it.”

Another OEM getting to grips with driver-in-the-loop simulation was Alpine. Rémi Auman, expert leader in simulation for ride and handling, noted that his team is targeting three areas for 2026 (see image below): enhancing driver immersion – an ongoing challenge for all developers; increasing the use of sim-workbench services; and improving multibody models. For comfort studies, Auman stated that the auto maker’s models are quite mature, so the next steps will be to adjust the settings and, “maybe in the future, we’ll be able to achieve real-time simulation,” he said.

With the conference’s emphasis on data harnessing, Cosin Scientific Software, which has been a partner of VI-grade for more than 15 years, launched its fresh approach to tire data management. With an increasing amount of users, the company has run into scaling problems when distributing high-quality input data. Developing countries especially cannot afford high-quality input data. To address this, the company has created a marketplace with a digital rights management system where tire data providers can market their data through Cosin’s software on the platform.

Cosin MD Gerald Hofmann explained, “Up until now, tire data has been sold as data files. This means once the data is out your hands, then it’s out in the wild. You can’t build a business model on this, which is why this data is very expensive. By putting the data under a digital rights management scheme, then the data provider can sell it using pay-per-use models.”

According to Hofmann, the development of this tire data library was an unexpected but welcome development for visitors: “We are very excited to see how this will be adopted through workflows which are currently very static. We’re hoping to break this up by giving access to a variety of input data. Users can allocate budget for a library of tires to choose from during the development cycle.”

AI specialist JuliaHub exhibited its latest technology, Dyad 3.0, which combines scientific machine learning with Gen AI. According to JuliaHub head of business development Europe Michael Hoffman, its name comes from its dual capabilities in performing controls and system modeling.

Asked what the company’s USP is, Hoffmann emphasized that having a single AI agent that can replace multiple tools is a powerful offering: “When we started Dyad we had AI mind – it wasn’t an afterthought where we had to do a plug-in. This means that the agent knows our architecture, and it only has to deal with one language.”

As expected, Dyad was met with both excitement and skepticism at the HBK SPS.

“We feel when talking about smart testing, we can add value,” added Hoffmann. “You have the measurements and we have the people to get the mathematical models or to refine their simulation models. For example, our value proposition to Rebel Dynamics [a servo platform partner of VI-grade], is that they have a mathematical model of their system, but they also have measurements, and there are differences between the two. We could help them minimize those differences so they can achieve better control.”

Related news, rFpro develops Japanese test route digital twin for virtual vehicle development