AI in the Auto Industry: Software-Defined Vehicles and Beyond

From manufacturing to simulation, automotive AI is more prevalent than you may think.

Cars are changing quickly. Most vehicles today are controlled by a plethora of low-cost microcontrollers, but over the next few years this control may be consolidated into a single software platform. The increasing role of software in the auto industry has its own name: the software-defined vehicle, or SDR, is a vehicle that is primarily defined by what the software behind it can accomplish.

“Auto manufacturers are currently focusing innovation on software development,” says Jan Becker, CEO and co-founder of Apex.AI, a firm that develops software for self-driving, electric, and connected vehicles.

With the role of software increasing in the automotive world, so too is the role of artificial intelligence (AI). Autonomous driving is only one example of how software and AI are changing what it means to be an automotive engineer.

Too Many ECUs

The average car between 5 and 10 years old contains anywhere from 50 to a several hundred electronic control units (ECUs). These are networked via a CAN bus, which helps the devices connect without a host computer. Such a setup causes the car to have a very complicated architecture.

“Today, almost every function in a vehicle, from the radio to the device that keeps a car in a specific lane, is powered by a separate microcontroller,” says Becker. “This means you can’t easily update the whole vehicle. With a software-defined vehicle, manufacturers will enable a car with new capabilities after a single download.”

Becker says that ultimately, auto manufacturers want to introduce one or a few software platforms. These platforms will rely on AI to coordinate and optimize functions. Becker shares that some manufacturers, such as Toyota, are keeping software development in-house. Others, like DaimlerChrysler, are partnering with chipmakers, including NVIDIA.

NVIDIA is already developing such a platform, Drive Thor, for autonomous vehicles. Drive Thor will pack a wide array of functions, including parking, driver and occupant monitoring, and rear-seat entertainment, into one platform. Drive Thor is expected to be available for 2025 models. Geely-owned automaker ZEEKR expects to utilize Drive Thor in EVs it will produce in early 2025.

Drive Thor, NVIDIA’s next-generation computer for autonomous vehicles, offers up to 2,000 teraflops of performance. (Source: NVIDIA.)

Drive Thor, NVIDIA’s next-generation computer for autonomous vehicles, offers up to 2,000 teraflops of performance. (Source: NVIDIA.)

 NVIDIA’s effort on Drive Thor reveals two challenges for auto manufacturers. Traditional auto companies such as General Motors, headquartered in Detroit, and Ford, headquartered in Dearborn, Michigan, can’t recruit software engineers in large numbers away from the mild climate and multiple opportunities of Silicon Valley.

“It’s a significant challenge for them to recruit candidates away from consumer tech. This is a substantial move away from where they were decades ago, when the focus was on designing the vehicle and the core expertise making engines and pressing metal,” Becker says.

Auto manufacturers are also frustrated with tech companies’ development of different types of software for cars, such as Android Auto and Apple CarPlay. When outside software runs on the head unit, located in the car’s dashboard, manufacturers are giving up control over a vehicle’s user interface.

“Yet (such) software can save time, effort, and money. It can become a selling point, and offer opportunities for partnerships with tech companies,” says Becker.

Auto manufacturers who produce electric vehicles (EVs) may find it easier to rely on one or a few software platforms. Manufacturers who build fewer vehicles have more time to refine and market such software, or partner with tech companies focused on such projects.

Making AI Part of the Process

Auto manufacturers are utilizing AI in design and manufacturing processes in many ways. Examples include:

  • maximizing AI-powered additive printing of metals for auto parts. The process is still more expensive and complex than additive printing of plastics.
  • placing edge AI devices on the manufacturing floor. This helps workers and machines analyze possible issues with safety, quality, and operations.
  • using AI-powered systems, like pairing of computer vision with deep learning models, to spot abnormalities in small components.

AI has also helped auto manufacturers deal with supply chain disruptions related to COVID-19 and climate change. With algorithms that incorporate data from weather predictions and consumer demand, they’re using AI to determine how many vehicles to produce and what parts they need to order ahead.

Software developers such as Nauto, a Palo Alto-based firm, also employ AI to help manufacturers understand driver behavior. Nauto’s AI-based platforms analyze how much a driver is paying attention to the vehicle and the road. Its algorithms look at a driver’s level of drowsiness and indications of distraction. It pairs this data with how fast the vehicle is traveling and how much the driver is accelerating. The software then provides audible alerts to the driver when necessary.

How AI Helps Test Corner Cases

One of the most visible uses of AI is in the testing of autonomous vehicles. Auto manufacturers rely on AI to improve advanced driver assistance systems (ADAS), a major selling point for new vehicles. One of the ways AI is helping manufacturers is to account for corner cases. These are outlier incidents that might happen once in a driver’s lifetime.

“With powerful cloud computing and AI, auto manufacturers can generate and test many millions of scenarios with typical variances, from bright sunlight to freezing rain. This way, a manufacturer can ‘drive’ the same road a million times in parallel. They can change everything from the brightness of the sky to the slickness of the road,” says Becker.

Becker adds that AI helps auto manufacturers estimate what might happen in an incident with less predictable factors, like a bird hitting the windshield as other drivers navigate the road.

Auto manufacturers also need AI to ensure simulations of ADAS systems are accurate. Stuttgart-based Robo-Test, a software developer that tests automated and autonomous driving systems, says “not all driver assistance systems are safe.” The company cites examples such as one ADAS system failing to recognize a semitruck when it was positioned perpendicular to a vehicle as opposed to head-on, and grossly underestimating the distance between the driver’s vehicle and another car just ahead. As part of its tests, Robo-Test recreated real accidents involving autonomous vehicles in a simulator.

The errors of ADAS systems are an indicator that fully autonomous vehicles may still be far in the future. In the coming years, the auto industry may see more closures of autonomous vehicle companies like Argo AI, Ford and Volkswagen’s joint venture that shuttered in October.

Reviewing all of the ways AI is integrated into auto manufacturing confirms AI is needed on multiple levels, particularly as a check on itself. Whether a vehicle is powered solely by algorithms and sensors, or a combination of these and a human driver, AI gives manufacturers a multitude of ways to reduce costs and increase consumer confidence in safety and vehicle performance.