Where are we on the road to autonomous driving?

The current state-of-play of ADAS technologies according to two industry specialists.

Regulatory hurdles, semiconductor shortages, geopolitical disruption; these are just some of the challenges currently facing the automotive industry. To adapt, many OEMs are rerouting investment to develop and deploy advanced driver assistance systems (ADAS) that are equipped with hands-off and eyes-off driving technologies for short-term returns.

While ADAS may have begun as a premium feature in executive cars, increasing interest from both consumers and safety regulators has driven a surge in adoption across all vehicle classes. As such, there has been a whole host of innovation in this area of late, from cameras and control units to radar and lidar systems, GPS and mapping data, to Human-Machine Interfaces (HMIs) and connectivity

In this article, two leading developers of ADAS hardware, software and systems share their insights into the market and what they believe will be the next step on the path towards autonomous vehicles.

INCREASING COMPLEXITY AND AI

“With increasing levels of complexity in ADAS/AD systems, it is becoming nearly impossible to thoroughly test and validate designs with traditional methods,” says Emmanuel Follin, senior manager, product management at Ansys. The company’s Autonomous Vehicle Simulation software supports the development, testing and validation of safe automated driving technologies while saving time and money for OEMs.

He continues: “An example of this is physical road testing, which is extremely expensive and time-consuming. As a result of these obstacles, virtual validation via simulation is gaining more traction than ever before. Especially with higher levels of autonomy, trusted, high-fidelity simulation tools are becoming a mainstream component of the autonomous industry’s manufacturer’s toolbox.”

According to Follin, another major trend is incorporating AI into ADAS design. “AI can coordinate vehicle connectivity, allowing vehicle information to be shared via high-speed wireless networks,” he explains. “AI also facilitates ADAS and AV perception using large amounts of data intelligence that is collected by sensors during drive time. This information guides system safety by improving system perception, helping ADAS vehicles respond to a variety of driving scenarios. One more area where AI/machine learning is making a profound impact is in the annotation of real-world driving maps and ground truth sensor data – this leads to robust perception outcomes and enhanced safety.”

This insight is echoed by Suraj Gajendra, vice president of products and solutions of Arm’s Automotive Line of Business. Over the past three decades, Arm has become a global computing platform, with more than 70% of the world’s population using products based on the company’s technology. Together with its automotive partner ecosystem, Arm provides OEMs with the processor IP, tools and software solutions for automated driving and software-defined cars.

“AI has unlocked new capabilities for every industry, and we’re seeing its impact on ADAS and autonomous driving,” Gajendra says. “Driver-assistance packages incorporating AI are helping to improve road safety and the autonomous driving experience, but they also require significant compute and performance. These demands are becoming more complex and costly to address with monolithic chips, which has sparked the industry to explore new approaches to building silicon such as chiplets. Chiplets allow the stacking of multiple semiconductor dies, leading to denser silicon design that can increase performance and lower power consumption. This approach is helping address the unique and specific challenges of the automotive industry, enabling more innovative ADAS features.”

INFLUENCING BEHAVIOUR

According to a study by the National Highway Traffic Safety Administration, an estimated 94% of serious car accidents are caused by human error. With statistics like this in mind, it’s important to consider how eyes-off and hands-off technology could potentially influence driver behaviour in the future.

“Safety is the main purpose of ADAS software – it was developed to help reduce traffic accidents through automated detection, navigation and avoidance features,” Follin says. “ADAS helps improve a vehicle’s awareness of its environment and aids its ability to respond to environmental factors within milliseconds, which helps heighten driver safety. Sensor technology (lidar, radar and cameras) plays a huge role in this, as different sensors work together to inform the vehicle of outside factors like pedestrians crossing the street, other vehicles occupying blind spots, and lane departures. ADAS can reduce driver stress because of the way it improves comfort and convenience and responds to outside variables.”

AVxcelerate Autonomy, Ansys’ end-to-end safety-driven toolchain, embodies this approach, combining statistics and simulation at scale to perform sensitivity and reliability analysis that is critical for the development of ADAS functions. Essentially, the platform enables car manufacturers to develop safe workflows for SAE L2/L2+/L3 sign-off processes and homologation. Leveraging expansive scenario variation management capabilities combined with physics-based sensor models, the toolchain ensures the accuracy and realism of simulations in order to effectively test autonomous systems.

“Advanced in-vehicle technology brings many positive benefits to driver safety,” Gajendra agrees. “Advanced technology in ADAS systems – some of which are already in vehicles today such as automatic braking or lane-keep assist – can help make decisions to manoeuvre the vehicle a certain way on behalf of the driver to prevent an accident. As this technology becomes more advanced, we will see better accuracy in the alert systems and ADAS decision making.”

DEVELOPMENT CHALLENGES

Despite the rapidly increasing capabilities of today’s ADAS systems, plenty of challenges remain regarding their development. “The growing sophistication of ADAS and autonomous driving features means greater code complexity,” continues Gajendra. “Today’s advanced vehicles already require 100 million lines of code, and it’s expected that fully autonomous vehicles will have up to a billion lines of code. To meet these growing software challenges, at Arm we are working with the automotive industry through collaborative initiatives like SOAFEE and the Autonomous Vehicle computing Consortium to enable open-standards architecture and accelerate automated and assisted driving systems.”

Arm recently introduced new technologies to accelerate vehicle development cycles while powering important functions throughout the car, including autonomous driving and critical real-time safety features. The company has also deployed its leading-edge Armv9 and Neoverse V3AE technologies to the automotive sector with the aim of bringing “server-class performance” to autonomous and ADAS workloads.

“In addition to the software, ADAS requires performant, power-efficient hardware for the software to run on top of,” he adds. “Most importantly, this compute must have safety built in from the ground up that meets stringent safety certification requirements. It will be crucial for automakers to ensure their software foundation is built on safe, power efficient and performant hardware to keep up with the ongoing technological advancements of autonomous driving.”

According to Follin, however, navigating the complex rules around regulatory compliance is perhaps the biggest challenge for manufacturers. “There is an exponential increase in complexity in ADAS/AD systems with rising levels of autonomy,” he says. “Obtaining regulation compliance for an SAE L2+, L3 level autonomy system is a lengthy, complex and extremely costly process to endure for manufacturers. It has been estimated that around eight billion miles of road testing are required to meet the necessary safety compliance. To achieve such an astronomical number with real world driving is a nearly impossible task and demands a sophisticated solution. Additionally, establishing the highest levels of safety is dependent on evaluating millions of unforeseen, critical edge cases that mimic real-world driving scenarios that can be potentially hazardous to drivers.”

ACCELERATING PROGRESS

With all this in mind, what are the next steps on the path towards fully autonomous vehicles?

“As AI and software requirements of software-defined vehicles increase, so does the need to get to market with technology solutions much faster,” offers Gajendra. “To do this, our partners need specialised silicon that allows them to power modern vehicle applications and do so without increasing costs or time to market. We are continuing to enable our partners with new types of silicon, including the Arm Compute Subsystems (CCS) for automotive, which we expect to deliver in 2025 to help reduce the risk of designing specialised silicon while allowing the industry to hit the performance and power requirements of today’s vehicle applications.”

According to Follin, “achieving higher levels of autonomy will require enhancements to existing features and the integration of sensor fusion techniques, greater reliance on AI-driven decision-making, and robust communications capabilities like Vehicle-to-Everything (V2X). With the collaboration of engineers, automotive manufacturers, technology developers and drivers, ADAS will continue improving road safety and the overall driving experience until fully autonomous driving becomes a reality.”