The project is a joint venture between connected road development company Cavnue and the state of Michigan (shaped like an oven mitt, stuffed up into Canada, David’s home) and the goal is to create a special corridor along Michigan Avenue and I-94 connecting Detroit and Ann Arbor that will allow for automated, self-driving connected vehicles of all kinds – mass transit, cargo, personal, whatever. Here’s how the State of Michigan describes the project: Ford is a major partner in this new venture as well, which does set them a bit apart from other major automakers that are working on automated driving tech like Tesla or GM. Ford CEO Jim Farley announced late last year the creation of a division to focus on AV tech and other futuristic stuff called Ford Next, and the VP of new business for that division, Franck Louis-Victor, gave this almost meaningless statement about the venture: The only thing to really take away from that heavily PR-massaged statement is this part: “potential to help accelerate the pace at which we can advance driver assistance features and safely deploy autonomous vehicles.” That’s because if we’re being absolutely, ruthlessly honest about when we can reasonably expect fully automated, self-driving vehicles (Level 5 on the SAE’s confusing scale) are really quite a long way away, in large part because reality is, let’s be honest, a bit of a shitshow. The Neural networks that underpin many automated and semi-automated driving systems, like Tesla’s Full Self-Driving Beta, suffer from something that has been called “brittleness” by researchers, which basically means (in the case of computer vision systems) that when an AI “sees” and “recognizes” something, another instance of that same something (road conditions or markings or signs or objects or vehicles) may not be properly identified because of what, to a human, would be a relatively small change. This means that for these visual AI systems to really learn to be fully independent, they need to have massive amounts of examples to pull from, and even then there’s no guarantee the next whatever will be recognized, because, again, reality is a confusing, chaotic place. It’s part of why we’ve seen these systems confusing the moon for traffic lights or stopping for stop signs on billboards. All of this is to say that if you drop down one SAE level, to L4, which is self-driving in restricted areas, then things get vastly easier, because if you’re only permitting automated driving in a contained area, then you can effectively control the reality in that area. That’s what would happen in this Detroit-Ann Arbor corridor: conditions would be made favorable for AVs, by minimizing the sort of traffic situations that would be encountered and providing more information to the cars from the road infrastructure itself. I think this approach will get us to fully automated driving a hell of a lot faster, and would be crucial in taking care of what I think is automated driving’s biggest unsolved problem: how do you deal with a disabled autonomous vehicle, as in how do you get it out of an active traffic lane if something goes wrong? If the car needs a human to take over, but the human is not responding, most current AVs and semi-AVS just stop in their lane. This isn’t a good solution, but it’s really all that can be expected – after all, if the car needed a human to take over, it’s because it was no longer capable of driving on its own, so how can it pull itself off the road? A special AV-friendly connected road could solve this problem. I even wrote about this very idea a few years back and my solution involved a sort of smart road as well. There’s a lot of advantages to letting the infrastructure help cars drive, but it’s not a solution everyone likes, by a long shot. Especially people with Silicon Valley tech backgrounds. This article about the new planned smart corridor on Forbes illustrates the other view well. The headline even comes out and says “A Dumb Highway Is Better.” The article suggests that the main reason a partially infrastructure-based solution is proposed is that infrastructure development is simply what the proposers know, and that the internet is a better model: The article also argues that any infrastructure development is slow and likely to be obsolete when finally deployed, and uses smartphones as another analogy as to why the smart car/dumb road model makes more sense: I have a lot of problems with these takes. First, and it’s weird that I need to even say this, but cars aren’t phones. You can take so many more risks with phones because when they crash, you just have to reboot them, as opposed to when your car crashes you need a paramedic to drag you out, potentially. In robocars, a “field” upgrade isn’t really a field upgrade. These vehicles can deliver themselves to service depots on demand to get their upgrades, unlike any other hardware product deployed into the field.” And as far as the “field” upgrades go, yes, Tesla did upgrade the computers in customers’ cars that wanted to use FSD Beta, but it cost $1,500 and I don’t see how updating tech on many individual cars is necessarily more efficient than updating tech on one roadway or network that propagates to all the devices that use it. Also, the cars driving themselves to the “field upgrade” depots thing, that’s not a thing yet, for anyone. And is the author saying that robotaxis will wear out a lot quicker? I mean, if they’re driving constantly, I suppose that’s true, though I’m not sure that’s a positive? I think we’ve seen over history that infrastructure networks actually can prove to be surprisingly adaptable to new technology. Hell, look at the old copper-line phone network, that went from purely analog telephony to digital traffic over the 20th century from wirephotos to telexes to faxes to modems to DSL lines, and there’s still advances being made to utilize this very, very old network. We use the same electric-power delivery network that our grandmas plugged their Electroluxes into to power and recharge all of our devices, including electric cars. It’s possible to design infrastructures in ways that are flexible enough to remain relevant even as the technology of the machines that use that infrastructure change. Yes, there’s still arguments about how difficult it is to get infrastructure projects to happen, since they’re tied up with politics and voting and people and all that mess, but the truth is that if we, as a society, are serious about coming up with a safer and better way to transport ourselves that uses automated driving tech, there’s really no great reason not to get the road infrastructure to help as an active partner in the process. I get the excitement and novelty of a robotic car that can drive itself anywhere, anytime, in any conditions, but that’s maybe too demanding a goal, and more importantly, maybe not even a goal that actually matters if we have a good network of Level 4-ready roads to use. If you want to drive anywhere with no restrictions, you can always use the squishy wet computer you have jammed in your skull. We’ve already seen that work at least reasonably well for over a century. I never expected full level 5 autonomy, and still don’t. https://youtu.be/F2iRDYnzwtk That’s all I really want too. Let me be able to take a nap on the highway I’ll be on for the next 8 hours, that’s all. I can drive the rest of the time, it’s fine. I swear, sometimes silicon valley types function by not understanding how anything works and then making a solution that takes a solved problem and deploys a worse solution. I’m sure I’m missing something, but I don’t understand why the autonomous driving folks wouldn’t be all over any incremental improvement to the viability of AV adoption. Smart roads to take some of the burden off smart vehicles even if only to increase redundancy and decrease liability? Intra vehicle communication to further strengthen the network and leverage more (and again more redundant) data? Hell, someone needs to develop the smart road tech, so why wouldn’t they want to be involved there? Unless it’s just a case of not-invented-hereness. We did not push forward in that direction, so infrastructure that would help everything communicate is now scoffed at. And that’s not even considering infrastructure and public transit that would reduce the number of drivers, which is almost entirely ignored, even as we consider not actually driving our cars. https://www.cringely.com/2016/08/25/self-driving-car-old-enough-drink-drive/ If anything, your example shows that infrastructure gets ahead of things. Interestingly, the 2-3 year update/upgrade cycle combined with infrastructure is a pretty solid way to progress tech (it’s pretty much the cell phone cycle). You keep improving the technology on both sides, without getting out of sync. Sure, it’s a lot of replaced computers, but not actually more than if you’re trying to make cars keep up with advancing automation tech individually. Aside from that, we can progress infrastructure without making existing tech obsolete. The choice to create phones without replaceable modems/antennae and the choice to increase data bandwidth by using different signals were choices made by companies taking into account upgrade cycles, consumer use, etc. Of course, the better solution would be good public transit for those who don’t want to drive. Cars could be for people who wish to drive, while the roads would get less congested when others use buses, subways, rail, or whatever else. But that’s just not the way it’s going. As opposed to the trillions of venture capital money flooding into the autonomous vehicle space, there’s fuck-all VC money to be farmed through infrastructure development. So the moment we as a society realize that the comfort, speed, and safety of transportation is an infrastructural problem, that VC money is drying up faster than…I dunno, some other thing that dries up quickly. And when that happens, that’s a lot of highly-paid consultants, executives, and machine learning PhDs without a gig. Thanks again for attending Daaan Raaants About Infrastructure Week here in the Autopian comments. Like Lake Mead. And like the bottom of the now dry Lake Mead you won’t like what’s there. “And when that happens, that’s a lot of highly-paid consultants, executives, and machine learning PhDs without a gig.” Kinda but not quite. As long as the car-road interface is kept relatively simple and robust this is really a no brainer. Less glamorous, but easily workable with current technology, likely relatively inexpensive to retrofit to existing vehicles (speeding up adaptation) and easily updatable centrally in years to come. The cell phone infrastructure is an excellent example of this approach. Now – and I’m only half kidding on this – they just need to adopt a failsafe financial model: Make it a federal project, conducted on interstates in all states at the same time, with jobs and money flowing to all places making it virtually impossible to cancel. Doing this, even before all development is done, is not the cheapest way to go. But it is certainly the fastest, and a way to make it the project visually and mentally connect with a large enough group op citizens that it will eventually succeed giving us at least something. And it would also force a resolution on the liability and insurance questions associated with automated vehicular transportation. Now that we have the ability to have semi-autonomous cars, we will probably spend the next 50 years creating new advances in engineering and regulation to make them work, including “smart” roads When comparing the current road network to the one in 1910, I think we have already engineered and regulated a lot of “smart-ness” into it (just not the type of intelligence that a tech-bro would acknowledge, because it isn’t based in silicon). Elon Musk and his ilk sound (to me) like a 1902 automaker bragging about how their car will soon be able to quickly and easily traverse muddy, cart-rutted dirt roads once they solve the problem of their skinny rubber tires blowing their tubes and getting stuck, when the actual solution was the creation of an intelligently planned and regulated network of paved roads. The real challenge for self-driving is the same as it was for all driving in 1902: a road network designed around the a new form of transportation whose needs are different than the forms that came before it. Myth of the Old West with the “rugged individualism” and the image of a white cowboy riding alone on his horse and blazing his own trail. (read: current Tech Bro in “autonomous Ford Mustang Mach-E”) Myth of the Commons with the “all in this together socialism” viewpoint with the image of the travelers on the well-rutted Mormon Trail in covered wagon trains that spanned thousands of miles. (read: current infrastructure using “smart highways and ubiquitous charging stations”) As with all things in this country, we live in the muddled middle. The pendulum will swing from one side to other depending on who holds the current power wand. (read: who is most wealthy) Warning! watch out for THE richest man in the world right now who holds the wand. (read E. Musk) Dammit Jason
