Covering All Bases: Using AI For Next-Level Automotive Testing

 

While modern cars may still look like cars, they are also now, at least in one sense, smartphones on wheels. Today’s automobiles come with many sensors, Driver Assist technology, and a powerful infotainment system. Moreover, the trend towards software-defined vehicles means mechanics and hardware are increasingly translated into software, wherever possible.

All these computer-driven functionalities constantly interact, crunching data from sensors monitoring the vehicle, its surroundings, and its passengers.

Buckling under the Weight 

Building this high-tech infrastructure is not enough. It must also be thoroughly tested to guarantee vehicle safety and comply with regulations.

Traditional testing methods buckle under the weight of the computerized car. Imagine testing the navigation system of a car driving on a multi-lane highway during a thunderstorm with voice commands issued by a distracted driver.

A staggering number of test cases must be generated to cover all possible scenarios on a vehicle platform that enables hands-free driving, lane centering, adaptive cruise control, smartphone connectivity, and voice interaction. Additionally, the platform must process input from LiDAR, radar, cameras, and vehicle behavior sensors.

Conventional Testing

The conventional but costly and time-consuming method of automotive testing is Hardware-in-the-loop (HIL) testing. It includes bench testing, manual test scenario generation, and physical prototype construction.

These methods also allow for limited coverage of so-called edge cases or situations at the limits of a car’s operating capabilities. In short, modern vehicles are growing too complex to rely on conventional testing methods.

With fierce competition in the automotive industry, time-to-market is crucial and increasingly complex. Hence, resource-intensive testing is becoming a major roadblock for innovation.

Kishore Raj, a US-based Akkodis consultant and artificial intelligence (AI) expert, has experienced the dilemma first-hand. He worked with an engineering team, collaborating with clients on automotive testing.

“Setting up the hardware, driving the prototype vehicles, and collecting the data takes a long time and requires a lot of manpower,” Raj said.

He and his AI team were confident that Generative AI could offer an alternative to conventional testing and approached a client, a leading US-based vehicle manufacturer, to present their ideas.

Speeding up Development

The client wanted to speed up the development of its ADAS (Advanced Driver Assistance System) and infotainment system and was looking for ways to reduce testing cycles and expand test case coverage.

The ultimate goal was to deliver better and more reliable new products than competitors by improving the system's reliability, performance, and scalability for future releases. At the same time, the team aimed to ensure compliance with global safety and regulatory standards.

“Manual testing wasn’t just slow; it missed critical edge cases,” Raj said. “And the cost of physical prototypes was rising significantly. We knew a new approach was needed, which could both broaden the scope of testing and deepen its precision.”

Akkodis proposed a Generative AI-based solution to automate and optimize HIL testing for the client’s ADAS and infotainment systems. The solution focused on two steps: first, taking data generated by the manufacturer’s own vehicles, applying GenAI models to analyze it, and generating edge cases from it.

This approach still did not produce all the scenarios that needed testing, so the second step used GenAI models to synthetically generate additional edge cases. This two-sided method led to almost 95 percent coverage of all possible test cases—a vast improvement compared to the less than 40 percent coverage in the initial test set-up.

The interface between the human and the vehicle will evolve. We must develop new GenAI-powered test systems to test this emerging human-in-the-loop experience level.

Kishore Raj, Consultant and AI Expert, Akkodis USA

Proof-of-Concept

The client initially commissioned Akkodis to build a proof-of-concept. Raj and his team subsequently developed a complete GenAI-based system for HIL testing of the client’s ADAS and infotainment system.

“We weren’t just testing individual components anymore,” Raj said. “We were able to test the entire ecosystem of the car’s infotainment and ADAS systems, replicating real-world driving environments with incredible detail. It’s one thing to test a navigation system on a bench. Still, it’s another thing to test while simulating complex edge cases such as lane closures, poor weather conditions, and driver inattentiveness. That’s the level of complexity we were dealing with, and GenAI made it possible.”

Reduced Testing Time

The new system reduced testing time by 40 percent, increased edge case coverage by 70 percent, and delivered a 30 percent cost reduction.

Implementing GenAI has reduced the roadblock of testing to a manageable size. Speeding up testing cycles from weeks or months to days has enabled the vehicle manufacturer to launch new features 25 percent faster than before. Moreover, the new approach has brought to light issues that would otherwise have gone undetected. If left unresolved, they would have led to software patches post-launch or even costly recalls.

Following this initial project, Akkodis has now been commissioned to test the client’s entire vehicle platform using these new methods.

Human-in-the-Loop

The next frontier of automotive testing will probably be simulating humans, Raj said.

“Vehicle manufacturers are increasingly looking into the interaction between the car and its driver and passengers.”

Infotainment and Driver Assist technology have already been adapting to the humans inside the vehicle. Cars can sense when the driver isn’t paying enough attention to the road by analyzing images from onboard cameras.

“We’ll see much more of that, and there will be a continuous loop between the human and the vehicle,” Raj said. “The interface between the two will evolve. We must develop new GenAI-powered test systems to test this emerging human-in-the-loop experience level.”