Mcity and the Drive Toward Autonomy

Huei Peng is no stranger to automating vehicles. Now he’s directing one of the places that are advancing the capability to realize self-driving cars.

Back in 1992, there probably weren’t a whole lot of people who were thinking about things like automated vehicles, but Huei Peng, who was at the University of California, Berkeley, at the time, submitted a thesis for his Ph.D. titled: “Vehicle Lateral Control for Highway Automation,” which is about, in part, developing the parameters necessary to assure that a vehicle will track down the center of its lane.

Fast forward 25 years and 2,300 miles, and Dr. Huei Peng is now a professor in the department of Mechanical Engineering at the University of Michigan, the Roger L. McCarthy Professor of Mechanical Engineering, an honor bestowed upon Peng for his “seminal research contributions in adaptive control and optimal control, with an emphasis on their applications to vehicular and transportation systems.”

You might also like

He also happens to be the director of Mcity (mcity.umich.edu), the world’s first purpose-built facility for testing connected and automated vehicles, a private-public partnership established in 2013 by the University of Michigan, in partnership with the Michigan Department of Transportation, on 32 acres of property. In November 2015 Ford became the first OEM to test at Mcity. Presently, OEM members of the facility’s Leadership Circle include, in addition to Ford, BMW, GM, Honda, Nissan, and Toyota.

Peng points out that the reason for something like Mcity for the development (early development; its sister operation, the American Center for Mobility (acmwillowrun.org), which operates on 500 acres about 17 miles to the southeast of Mcity, is better suited to perform more production-ready advanced drive assistance systems (ADAS) developments given its size and scope) of ADAS is because “it is a controlled test environment.”

Peng explains, “A few years ago, people thought we would test the automated vehicles on populated roads”—public roads. “First, it is not efficient, and second, it is not safe. It is not efficient because you’re not doing a controlled experiment. You are running into interesting traffic scenarios. If there is a problem, issue or challenge, you’re going to want to do that again and again, repeatedly, trying out different control settings and algorithms, but you can’t because it is not controlled.” It is something that just happens and is then gone. “It is not safe,” he adds, “because it is using the public roads.”

Remember that they’re doing early development work at Mcity, work for which having unwitting, unknowing participants would probably not be a good thing.

One of the things that Peng and his Mcity colleagues think is key to the development of fully autonomous vehicles is communication. Communication from vehicle to vehicle (which is designated “V2V” among those who talk about it so often that they need to make an acronym out of it) and vehicle to the environment or infrastructure (this is V2X; for example, the traffic signals at Mcity communicate with the vehicles running routes ).

Peng says that cars equipped with vehicle-to-infrastructure capabilities “give you day-one benefits.” He explains, “If I buy a car with V2V today in the U.S., there is only one, the Cadillac CTS, that has it.” In other words, the 2017 CTS can talk with other 2017 CTSes. Toyota offers five models with V2V—in Japan. But when it comes to V2X, there is an increasing amount of communications infrastructure being built out by governments (e.g., the state of Michigan). So every car with V2X capability can benefit.

“We believe connectivity is essential for automated vehicles,” Peng says. Why? For one thing, it has a longer range and doesn’t require line-of-sight like a sensor (radar, LiDAR or camera). And the information is much richer, not a set of, say pixels that have to be analyzed. Peng points out that through communication a vehicle can “tell” other vehicles where it is, its velocity, acceleration yaw rate, steering angle, weight, whether ABS has been activated. . . and more. “It is a much richer set of information than can be provided by onboard sensors.”

This isn’t an either-or situation, however.

A question about all this, of course, is simply why? Why is there interest in autonomous driving capability? With all of the emphasis by virtually every OEM (and not just the ones who are in the Mcity Leadership Circle) and major Tier One suppliers, insurance companies, telecommunications companies, etc., it seems that autonomy is an inevitable foregone conclusion. But why?

Peng says that it is first and foremost about safety. There are, he explains, on the order of 37,000 lives lost on the roads in the U.S. annually and 1.2-million on a global basis. Human factors are instrumental in about 94 percent of all of these instances. So the notion is that by having automation—a robot—driving the vehicles, there can be safer trips.

Peng admits, “Robots have their own problems,” then quickly adds, “But they keep improving.”

He uses an analogy: chess-playing software. Early on, the human player was better than the computer. But the computer kept improving. Every time it played it learned something. So with time, it got better and better. That is, in February 1996 Garry Kasparov, the chess champion, beat IBM’s Deep Blue. In May 1997, Deep Blue won the match. And that’s pretty much been the case ever since, as the computer keeps learning and learning. This is similar to the computers that are being deployed in vehicles (and which goes back to Peng’s point about the ability to repeatedly test events and conditions at Mcity).

In addition to which, robots can be programmed to abide by the law, which can result in better traffic flows.

A second reason why autonomy can be beneficial is that it provides access to mobility to those for whom it is difficult or impossible to drive, such as seniors and the handicapped.

But the number-one challenge at this point is the cost of equipped vehicles with sensors, which at this point are still prohibitively expensive so far as many consumers are concerned.

And right after that is performance: “Understanding the typical driving scenario is relatively figured out,” Peng says, “But understanding the one-off situations—the edge cases—like a flooded road, a big pothole or a policeman waving at you can still be a problem.”

That said, Peng thinks that autonomous vehicles will make their way onto the roads, but probably in fleet applications.

That is, he points out that there are a variety of things—like dirt and accidentally knocking a sensor out of alignment—that can have a negative impact on ADAS. “Is a car company going to be willing to sell a vehicle to John Doe who never washes his car or who will bump the camera and not go to the dealer for recalibration?” Chances are better that initially, professionally managed fleets, where car washes and system adjustments are regular events, make more sense.

What’s more, there are conditions—be they weather or the prevailing environment—that may make fully automated driving extremely difficult to realize. “When you can deal with the cows in Mumbai, the kangaroos in Australia and the heavy snows in Alaska, then there will be Level 5 automation,” Peng says, adding, “But I don’t think we’ll be there anytime soon.”

But it won’t be because of a lack of effort at Mcity.