You might think that as the senior vice president, Automated Driving, Toyota Research Institute (TRI), the billion-dollar organization established in 2016 to focus on artificial intelligence (AI) and robotics particularly as they apply to transportation and mobility, Dr. Ryan Eustice would be all about getting the human driver out of the loop. But when asked about the SAE levels of autonomous driving and how the work that TRI is doing fits into it, Eustice says, “The thing about the SAE levels and the thinking behind automation is the mindset of taking degrees of freedom of control away from the human and handing it over to autonomy.” That is, as you go from 2 to 3 to 4 to ultimately 5, there is less responsibility for the driver and more for the automation.
“What’s different, and how we flip things around is asking ‘How about if we have the AI guarding the human?’”
In other words, the human is still part of the system—is the key part of the system—but is augmented by sensors, actuators and controls. However, unlike some sort of bionic driver, someone who is visibility enhanced by all manner of advanced technologies and so goes cutting through traffic like an android, the TRI approach is far more unobvious.
Rather than following the SAE levels (“What we think about is a perpendicular axis to the levels,” says Dr. Gill Pratt, TRI CEO), they have two distinct approaches to automated driving. There’s Guardian. There’s Chauffeur. Both are to do what their names imply, with the Guardian serving the enhancement and protection functions and the Chauffeur being the self-driving technology that will allow the driver to be a passenger, in effect, either under all conditions or under restricted circumstances (predicated, say, on a known geography and appropriate weather conditions).
Guardian for All
Perhaps because this is Toyota, these two tracks aren’t separate inasmuch as Guardian is a system that not only can work with a human but with an automated system, as well. (And in that regard, TRI is going to make Guardian, in some form, available to other OEMs or developers of autonomous driving systems. As Pratt said in comments at the 2019 CES, “We believe in it so much, we would like to see it on every car on the road, not just Toyotas. In other words, Guardian for all.”)
What’s more, Eustice notes that they are working “to build the Guardian and Chauffer systems in one unified technology stack,” such that the systems stretch from one side of the automated driving spectrum to the other. Or, as Pratt puts it with something of a fundamental spin on what is an incredibly complex technical undertaking, “TRI is trying to build a generic system that has a bunch of knobs on it. And we can tune those knobs depending on how good the sensors are and how many there are.”
Realize that they are committed to developing fully automated driving. But Pratt points out, “Level 4”—the fully autonomous capability within parameters—“is going to take a long time to spread. What can we do that’s good for society with the same technology”—as in lidar, radar, optical cameras, thermal imaging, ultrasonics, advanced processors, robust software, and actuators—“in the near term?”
And the answer to that question is Guardian.
Eustice explains, “When you think of Chauffeur, the car is doing the driving task. In the operational design domain the vehicle has to drive 100 percent of the time. The opportunity with Guardian is that the human is driving but using the same core technology that goes into Chauffeur, but the system is evaluating how confident it is.”
To put this into some sort of understandable context, think of a slalom course that’s made up of a sequence of orange cones positioned so that in order to maneuver through them, the steering must be done in a precise manner. A human driver, a good one, goes through and knocks down a few cones.
Then the same driver in the same vehicle goes through but this time Guardian is engaged. The driver is still in control of the steering wheel and the pedals, but in this case the system provides inputs such that the vehicle is able to thread its way through the cones without taking any of them. It is not a matter of the steering wheel being jerked out of the driver’s hands and the pedals acting as if they have minds of their own. Because these are by-wire systems, controlled inputs can make the required adjustments. It is more subtle, blended, leaving the driver feeling like a far better driver than is actually the case.
But there is something that Pratt says about the Guardian system that is on the one hand surprising because people in the auto industry don’t ordinarily make such admissions and satisfying because when it comes to something that has to do with safety, clarity is better than obfuscation: “Guardian doesn’t guarantee that there will be no crash. That’s impossible to do. If you’re boxed in and suddenly a car comes in from the side, there’s nowhere to go and you’re going to be in a crash.” Physics are physics.
What’s more, the system can determine whether it has the bandwidth to do something or if this goes back to the responsibility of the driver.
Another aspect that the researchers have to take into account is something that developers of any automotive part, system or entire vehicle must always be aware of: cost. “We must be conscious of the cost to deploy,” Pratt says. He amplifies: “Cost is the number-one issue.”
As there are advancements in things like sensors at a fast pace, a determining factor regarding their applicability to a vehicle is not the cost of producing a few thousand of them, but what they’ll cost when the volumes reach the millions: “Is the physics of the device, the engineering, such that it could reach a lower cost when it is made in car volumes?”
“This is a challenge for every startup company we see: Could you eventually do it at this cost?”
He recognizes that the margins in the auto industry are slim and that the amount of additional cost that can be added to the MSRP for a personally owned vehicle is small (versus a fleet vehicle that would be operated in an entirely different regime, thereby changing the economic justification). “This is a challenge for every startup company we see: Could you eventually do it at this cost?” Economies of scale still matter.
Eustice raises another issue: “Getting to automotive grade is also a big deal.” It is one thing to have a sensor that works well under test conditions, but something else entirely in the “real” world: “Cars today,” Eustace points out, “have backup cameras. How many times have you had to use your thumb to clean the lens?” That thumb approach isn’t going to be particularly efficient when there is an entire sensor suite.
And there is another issue, one that is huge given the complexity of an automated driving system: it must be simple to use. As Eustice puts it, “Ultimately, I don’t want my grandmother to get in the car and have to read a manual to understand what Guardian is.”
Continental, an automotive supplier that has a deep engineering bench, is making a huge organizational change, one that Dr. Elmar Degenhart, chairman of the executive board, explains is necessary because, as he puts it, “The industry is changing at a high pace, so we have to change, too.”
According to Frank Jourdan, president, Chassis & Safety Div., Continental Contitech AG (continental-corporation.com), the high-resolution 3D flash LIDAR (HFL) technology that the company is developing for deployment in automated driving systems in the 2020+ timeframe provides an array of benefits.
Visteon Corp. is developing DriveCore, an open platform to control and operate autonomous vehicles.