While companies such as Google chase the fully autonomous car, Toyota is taking a more measured approach toward a “guardian angel” car that would seize control only when an accident is imminent.

But as starkly different as those approaches are, they both will require a wide range of data-intensive technologies, according to Gill Pratt (pictured), chief executive officer of the Toyota Research Institute, a research center focused on AI and robotics. He spoke at the GPU Technology Conference in San Jose today.

Toyota has made a huge bet–a billion dollars over five years, in fact–not only on semiautonomous cars but robots that could help older people with indoor mobility. The Toyota Research Institute, which will have facilities near Stanford University and the Massachusetts Institute of Technology, is intended to focus both on what Toyota calls outdoor mobility (cars) as well as indoor mobility (robots).

Toyota was already facing plenty of competition in its core automobile market from the likes of Google and even Apple, as well as data-driven cars from Tesla. But in looking beyond cars themselves to the application of emerging vehicle technologies to other markets such as robotics, Toyota is opening the way to both huge opportunities and huge new competitors.

On its traditional auto front, Toyota’s emphasis on helping drivers rather than usurping them sets it well apart from companies such as Google and Baidu that are working on fully autonomous cars. But the philosophy also may hold more appeal both to drivers and government regulators in the short term. It also meshes with an increasing number of artificial intelligence researchers at IBM, Amazon and other companies who believe AI will do best in conjunction with human intelligence rather than trying to completely recreate human cognition.

Either way, developing these cars will require automakers to morph from hardware manufacturers into enterprise technology companies. The amount of computer power and data analysis needed to make the cars work looks to be staggering.

“For the first time, the car companies seem to know what they’re getting into,” Patrick Moorhead, president and principal analyst at market researcher Moor Insights & Strategy, said after Pratt’s keynote. “They realize the amount of data and processing that’s required to pull this off.”

But he said that cars literally will have to have a supercomputer inside to pull off what they’re trying to do. And that currently requires space, serious fans and water cooling that automakers aren’t yet ready to include in any but the most expensive cars. “A car can’t afford a supercomputer of 2017,” Moorhead said.

Even before those computers, one of them announced earlier this week by conference sponsor Nvidia, go commercial in a big way, automakers such as Toyota are already bringing a number of data-intensive technologies to bear on making cars safer, easier and less environmentally damaging to drive.

For one, Pratt is looking to leverage work he headed when he was a program manager at the Defense Advanced Research Projects Agency (DARPA), where he worked on robotics and neuromorphic computing. He views neuromorphic chips such as IBM’s TrueNorth brain-inspired chip as one way to vastly reduce the amount of power needed to do the massive computing tasks needed for cars of the future.

Pratt also is looking to apply some lessons learned from the DARPA Robotics Challenge, an annual contest to advance the state of robots, which he ran last year. The central lesson was that complete autonomy was very difficult to achieve successfully, suggesting that a better approach is “parallel autonomy.” That’s when a robot helps only when it becomes apparent the human is having a problem.

In cars, that translates to what Pratt called a “guardian angel” model, instead of the “chauffeur model” that fully autonomous cars represent. The latter, he said, can’t really be deployed until it works perfectly; consider the kerfuffle over the accident Google’s self-driving car recently caused even though no one was injured. It also puts all the liability on the manufacturer, the driver’s skills are ignored, and there’s a big handoff problem. That’s the disconnect when people who haven’t been paying attention to driving suddenly are asked by the car to take control.

In contrast, a “guardian angel” car that only kicks in when it looks like an accident could happen skirts many of those problems. Mainly, it can be deployed now to reduce traffic accidents, saving lives immediately, and there’s no handoff problem. “We feel this is a very valid approach,” Pratt said.

Toyota actually is pursuing both approaches, since the technologies involved are similar and one advantage of fully autonomous cars is that they can serve people who otherwise can’t drive. In fact, Pratt announced today that Toyota is opening a new site in Ann Arbor, MI, to house 60 people who will work on the chauffeur model. But it already had announced plans to hire 150 people for a lab near Stanford to work on the guardian angel approach.

Photo by Robert Hof