Going for Level 4 autonomy—where the car drives itself and you can go to sleep—is typically justified on the grounds that such cars will be very safe. And they had better be, or we’d never let them loose on the roads.
But the safety-first argument is flawed, says Gill Pratt, who heads up self-driving car research for Toyota. Reason: Safety can be obtained by other means.
“The reason for Level 4 being done—to save lives—is backwards thinking, even if you assume it’ll be 10 times safer,” he tells IEEE Spectrum. “That’s not the only way to save lives; there are multiple ways to do it.”
Pratt allows that there’s a purely economic argument for self-driving cars—remove the driver and you cut expenses in any commercial application, like taxi service and trucking. But that decides things only after self-driving tech can be proven far better than the best human driver. A system that’s just 10 percent better will win over statisticians and philosophers but not the general public.
This isn’t the first time Pratt has poured cold water on the idea that we’re on the verge of getting rid of the steering wheel and pedals, as GM Cruise plans to do in a pilot program next year. Read our Q&A with him from early last year. But nowadays, Pratt’s emphasizing how a system that is essentially Level 4 can be repurposed as a teammate to the driver, rather than a replacement.
Toyota is developing Level 4 systems, he said, but when they’re purposed to drive the car—and thus called Chauffeur—they need vastly more validation than has been done yet to be made into a generally useful product. Toyota doesn’t expect to hand a Level 4 Chauffeur to the public for years, though the company plans to demonstrate one during the 2020 Olympic Games, in Japan, within a relatively limited environment.
But what Pratt calls the “technological equivalent to Level 4” is coming much faster. It’s called Guardian, and he says it’s a lot better than today’s advanced driver assistance systems (ADAS), which offer lane keeping, active cruise control, and emergency braking. “We think Guardian features will trickle into production vehicles soon,” he says.
Here’s how it looks in practice:
Guardian uses a diversity of sensors and maps which, though they might be a little out of date, at least tell the system the most likely environment it’s in and the location of the car in that environment. A prediction system figures out how the environment around the car is likely to evolve, and then a planner works out the car’s trajectory and other behaviors.
“It asks if there’s an unprotected left-hand turn or a highway merge coming up,” he says. “When the system’s functioning as Guardian, it’s there to warn or nudge the driver, and if things are really bad, to take over temporarily.”
We already have a Level 2 system—the Super Cruise function, which is available in the Cadillac CT6. As Lawrence Ulrich reported in April, it’s the current self-driving champion of production cars. But to make sure that the driver doesn’t get lulled into dangerous complacency, the car uses cameras to observe the driver’s eyes and body posture and to jostle him or her back to situational awareness if necessary.
But Pratt suggests that approach reflects backward thinking, too.
“We’ve known since the 1940s that the better the autonomy, the more you tend to overtrust the system,” Pratt says. “That’s why Super Cruise has a monitor that watches you. In Guardian, we’ve flipped the whole nature of who guards whom: We have the person drive.”
Philip E. Ross is a senior editor at IEEE Spectrum. His interests include transportation, energy storage, AI, and the economic aspects of technology. He has a master's degree in international affairs from Columbia University and another, in journalism, from the University of Michigan.