AI Expert Says Autonomous Vehicle Systems Need Work

Several years ago, more industry observers thought the current decade would be the time when autonomous vehicle technology would steadily phase into many new vehicles and gain significant acceptance before the end of the decade. But in 2024, autonomous vehicles, while still on the radar of many vehicle manufacturers, appear all but relegated to being in niche applications. Ongoing concerns about AV safety, magnified in the light of several accidents in vehicles relying on autonomous driving systems, have forced manufacturers to scale back or delay AV development.

In a recent interview with Design News, Dr. Stefan Heck, CEO of Nauto, a vehicle AI technology provider, spoke about several incidents with autonomous driving systems in advanced vehicles.  For instance, last October, a woman was trapped underneath an autonomous vehicle after being hit by a human-driven car in San Francisco. The driver of the human-driven car ran off, while some experts acknowledged that the AV, while it did brake, was not likely able to stop in time to avoid hitting the woman.

“Let’s not take control away from humans,”  said Heck. Nauto has developed AI technology that still relies on driver input while accounting for many of the factors─weather, lighting, road surface, other vehicles, and pedestrians─that could impact vehicle safety.

Related:What Is the Future for Porsche Drivers in an Autonomous World?

“All the major automakers have reset their AV roadmaps,” said Heck. He noted that vehicle makers are incorporating some level of autonomous technology, such as parking assist or autonomous parking, rather than emphasize autonomous technologies intended to let cars drive themselves for any extended period.

Heck believes it will take at least two auto development cycles to see fully autonomous technologies in vehicles. The time frame for this would be the next decade.

One problem, Heck noted, is the wide range of incompatible systems in most vehicles. “Systems are built by separate suppliers and they do not communicate with one another.” While Heck does believe that software-defined vehicles (SDVs) will be an enabler of technologies that communicate with one another, in the near term the trend towards SDVs could slow autonomous vehicles as OEMs grapple with achieving a common interface.

“Within about five years, software-defined architectures will become a big game changer. More vehicle issues will be fixed not through hardware but through over-the-air updates.”

As far as the progress of AI algorithms, Heck said more is needed in visual AI algorithms, which are more critical in AV driving systems than say, generative AI which is based on large language models. “Visual algorithms are much more complex to interpret,” Heck noted.

Related:AI Takes Command of the Auto Industry

In addition, Heck said modeling AV systems in inherently complex because one has to consider the actions of other drivers. “What are other drivers thinking?” Heck added.

Heck believes that in the near term, AV technology will likely make stronger inroads in fixed-route applications, such as fixed-loop bus routes that operate on a predictable schedule in relatively controlled environments.