When a self-driving Uber car killed a pedestrian in Tempe, AZ, a few months ago, it shook up the mobility industry.
The accident served as a gut check, and it sent already tenuous rates of consumer acceptance plummeting. One of the main benefits associated with driverless cars is safety, but it’s hard for riders to feel safe if the threat of death is there.
Despite the leaps and bounds made in A.I., sensor, and computer vision technology in the past decade, teaching cars to think like humans and anticipate what will happen next turns out to be very difficult, and some in the industry are coming to the realization that quite a bit more maturation is necessary before these technologies will be ready to be used commercially at scale. In fact, at last week’s Intelligent Transportation Society of America conference in Detroit, we heard more than one company say they have doubled down on driver safety features that can be deployed in the near term while we wait for full autonomy to take root.
One such company taking a safety-first approach is Toyota. We had the opportunity to chat with Ryan Eustice, who heads up the Toyota Research Institute’s autonomous vehicle team in Ann Arbor, MI. He shared details about what TRI is working on in Michigan and elsewhere, Toyota’s approach to safety, the advantages to innovating inside a huge automotive company, and the value of TRI’s relationships with major research universities. Eustice says Toyota will hold the first public demonstration of its autonomous vehicle technologies at the Tokyo Olympics in 2020.
TRI was established in 2016 as a separate business unit of Toyota with an initial budget of $1 billion over five years. Its mission is to advance the development of autonomous vehicle technologies, with a focus on A.I. and machine learning, robotics, materials, and user experience. In addition to Ann Arbor, there are two other TRI locations in Cambridge, MA, and Los Altos, CA.
TRI’s business model involves “a vigorous association” with research universities, Eustice says; the company’s locations were chosen due to their proximity to MIT, Stanford, and the University of Michigan. Globally, TRI employs more than 300 people, about 50 of whom work in Ann Arbor.
“Toyota had the foresight to create TRI in the United States to give it a long runway to compare with Apple and Google,” Eustice says. “We’re more agile [than traditional automotive entities], but we also have the backing and commitment of a big OEM and a clear path to product commercialization.”
Eustice works simultaneously on the development of two different automated driving modes with identical underlying technology called Chauffer and Guardian. Chauffer mode is more of a traditional self-driving play, where humans are strictly passengers and the car is responsible for its operation. Guardian, on the other hand, allows humans to drive while A.I. monitors everything closely in the background. The car only intervenes and overrides the driver when it’s necessary to prevent the car from crashing.
“When you look at level 2 or level 3 autonomy, the human is in the back-up or supervisor role to the A.I.,” Eustice explains. “With Guardian, the A.I. supervises human drivers. We’re not taking humans out of the driving task, but making them better. Our view is unique to the industry—Guardian is a way different stance than anyone else I know is taking. Guardian doesn’t even fit SAE’s 0-5 autonomy scale, which is all about removing human drivers and offering convenience. With Guardian, we’re focused on safety and creating a non-crashable car that works with you.”
Eustice says industry-wide, replicating human decision-making is one of the biggest challenges yet to be solved: “A big part of planning is prediction, and what makes that hard is people.” For example, the car must predict whether the pedestrian about to cross the street will look down at her phone midway through the intersection and stop, or do something else that’s unexpected. “People are the source of a lot of driving uncertainty, and there are a lot of subtle cues to learn. When does green not mean go?”
The goal with Guardian, Eustice says, is to make it simple and user-friendly enough that his 80-year-old grandma could grasp it. To test the system, TRI is in the process of building its own closed track in Ann Arbor, which is expected to be finished by October, but it also tests its cars on public roads. Michigan’s industry-friendly regulations are one reason TRI is running tests there, but it also does testing in California and the Boston area.
Eustice splits his time between TRI and U-M, with 80 percent spent at TRI and 20
