Toyota's two-step autonomous push

DETROIT — First your Toyota will be your guardian. Then it will be your chauffeur.

Ryan Eustice, a newly hired autonomous driving guru at Toyota Motor Corp., says the company’s first truly self-driving cars will arrive in five to 10 years — in that manner and in that order.

The rollout is being handled through two programs established this year under the Japanese carmaker’s newly created, U.S.-based Toyota Research Institute Inc., a $1 billion effort charged with developing artificial intelligence and advanced robotics.

The first project, code-named Guardian, will develop next-generation driver-assist safety systems and is being led out of Toyota Research Institute’s laboratory in Cambridge, Mass. The second, project Chauffeur, aims to develop truly driverless cars out of the institute’s office in Ann Arbor, Mich.

Overseeing both are Ryan Eustice and Edwin Olson, two University of Michigan professors tapped this year to lead the world’s biggest automaker into autonomous driving.

The target is a so-called Level 4 autonomous vehicle — a car capable of driving itself anywhere, anytime, under any weather scenario, while the human sits back, hands off.

Toyota’s first applications of Level 4 autonomous driving will be “geofenced,” said Eustice, who, with Olson, is co-director of Toyota’s autonomous driving program. That means the cars will be fully self-driving but in restricted areas such as interstate freeways.

“In reality, I don’t think anybody is going to launch a product and say on day one it will be Level 4 everywhere,” Eustice said in an August interview. “That is a monumental goal.”

Geofenced Level 4 Toyotas, however, are just five to 10 years away, he said. 

Eustice, a Massachusetts Institute of Technology-trained robotics engineer, has been researching autonomous driving at the University of Michigan since 2008, long before he caught the attention of Toyota. The 40-year-old looks every bit the academic lab rat, from his shaggy hair and long sideburns to his polo shirt and faded bluejeans.

He joined Toyota Research Institute in March and opened its Ann Arbor office in June, bringing on board the small team that had been working with him on the technology at the university.

It hasn’t always been easy blending theoretical, ivory tower research with the market-driven demands of one of Japan Inc.’s most profitable companies.

“It’s definitely a learning process on both sides,” Eustice said. “The mindset of the Silicon Valley approach is to want to go fast, to be agile, to fail quickly. This is a unique benefit that TRI actually provides Toyota Motor Corp.”

Traditionally, a carmaker would work with different suppliers to source components and systems. But Toyota’s new model is to integrate that technology in-house, Eustice said.

“We approach this very much like a Google or an Apple, a large-scale modern software company … and I think this is new for Toyota,” he said.  “This is a new learning curve.

“TRI is set up to be its own stand-alone thing, with its own processes and procedures,” Eustice added. “But we have a very strong connection back to Toyota Motor Corp. The code that we’re writing now is actually going into production vehicles.”

Gill Pratt, CEO of Toyota Research Institute, said in June that advanced safety functions, such as those being developed through project Guardian, would see the quickest payoffs.

The breakthroughs, he said, likely will focus on evasive technologies that sense when a collision will happen and steer the car to safety. While such technologies are already on the market, Pratt suggested that next-generation versions may steer a car around obstacles even if it means steering out of the lane.

“We see Guardian as one [level of technology] we can bring earlier,” Eustice said. “That same common core of technology can support the Chauffeur application.”

By way of project Chauffeur, Toyota has said it will introduce semiautonomous vehicles for certain highway driving by 2020.

Those vehicles should be able to merge and exit highways, maintain and change lanes and keep a constant distance from other vehicles on the road, all without driver input.

Beyond that, some of the biggest hurdles are teaching the computer to think with the intuition and contextual understanding of the human brain, Eustice said. When, for example, does a green traffic light not mean go?

“It doesn’t mean go when there is a guy with a badge standing on the corner of the road holding his hand up,” Eustice said. A computer program can be written to detect a person in that pose. But a human brain weighs more context, such as whether the person is in uniform or whether other cars are stopping, to determine whether it’s a police officer or just someone pulling a prank.

“This general-purpose reasoning is one of the big challenges,” he said.

Another challenge: bringing Toyota Research Institute’s autonomous driving team up to full staff.

The Ann Arbor office has about 25 people now, and Eustice aims to increase that to 50 by the end of the year and eventually expand it to as many as 70. That would make Ann Arbor the institute’s second-biggest campus behind the headquarters in Palo Alto, Calif. That office has a couple dozen people now and is expected to grow to 150, Eustice said.