Will the pandemic map a new course for autonomous cars?
From where we sit, it looks much less like an economy that could benefit from autonomous, self-driving cars that wheel their snoozing occupants safely from place to place, than just three months ago. For that matter, we can probably scratch our heads now about whether that moon shot by 2024 is a great idea. Our priorities have been shifted for us.
But relatively few of us may be sitting in China right now. After the country's late March announcement that its hospitals were reporting no new coronavirus cases, it began lifting its stay-at-home orders, and launched an aggressive marketing campaign aimed at portraying its economy as having weathered the storm stronger. Perhaps nothing symbolizes a normal, healthy economy (even if it's more symbol than substance) than the resumption of the country's autonomous vehicle development programs. In late April, Hangzhou-based DAMO Academy -- funded by the country's B2B colossus Alibaba -- released the first footage from an extensive simulation platform launched as the stay-at-home orders were being lifted from Wuhan province. The tape appears to show a virtual, autonomous delivery vehicle navigating a virtual, busy, semi-organized city, full of jeeps that run stoplights and children darting out into streets.
It's a robotic delivery vehicle. It can't go very fast, but it has a mission to complete: probably delivering critical supplies to a metropolitan destination. You have to stop just for a moment and think about how useful such a vehicle would have been for all those urban residents stuck at home needing food and perhaps medicine.
Maybe those of us who concentrate on the consumer end of technology are sitting in the wrong place. For China, investments in technological infrastructure are made by the country as a civic priority, rather than by venture capitalists with an interest in successful commercial products. Maybe it's time for a little altruism -- for considering the benefits of autonomous transportation beyond taking a few commuters' hands off their wheels.
Also: Is this the tipping point for delivery by robot?
"We are going to be living in a world, because of climate change, with more extreme events. We know that," remarked Dr. Sherman Robinson, senior fellow with the Peterson Institute for International Economics. "We're seeing more worries about international supply chains in food and agriculture. A lot of that stuff goes through particular choke points in the transportation network, and you want to diversify."
There may be more benefits to transportation autonomy than just giving a driver a nap.
"A lot of these applications of 5G -- vehicular, factories -- are for automation," said Prof. Robert W. Heath, Jr. of the University of Texas at Austin -- one of V2X's most ardent and earliest supporters. "The application there is normally about getting humans out of the loop, and is maybe saving cost, or allowing more rapid reconfigurability -- things like that. But I think we also see now that automation has a big role to play during a pandemic, when you just want to avoid human contact. You're not trying to get rid of people, but you're trying to make everybody safe."
We don't need anything else driverless these days
I don't usually get personal with my technology infrastructure stories, but here I go anyway: For the duration of my career, I have been a skeptic of autonomous vehicles. Whenever I managed journalists covering the annual CES conference, I instructed them to avoid getting sucked in by the self-driving car demos. They're too much like the "Future World" exhibits at '60s theme parks, I told them: ideas packaged and presented for the sake of production, not substance.
Occasionally, I've been asked why the lack of interest from me, and why I think that disinterest extends to readers. I would respond that a technology is not worth covering if its benefit could not be explained in 12 words or fewer. ("You're such a hypocrite," someone once shot back at me, and they had a point.) "Automating society" is one benefit I was told, but I'm still not certain what that really means, besides the many times the concept was attempted in Doctor Who by the Master, the Daleks, and the Cybermen. Last year, in an effort to flesh out the technology's true purpose (but probably also to get me to respond to story pitches) the benefits of autonomous vehicles became associated with the broader, more urgent, mission to save our planet -- for instance, reducing carbon emissions and saving fuel.
I believe in technologies that assist and enable people, not replace them. I support robotics that boost people's capabilities to perform useful tasks and produce useful things. I support robots that can substitute for human beings in survivable scenarios, such as burning buildings and hostile planets. And I believe those critical and laudable carbon-related goals can be more readily achieved if we devote our attention to putting carbon to better use.
So I have more than a bit of difficulty accepting the notion that among the more urgent priorities that technologists face now -- especially now -- is that anything I can download from the web should replace me in the driver's seat of an automobile. The more compelling counter-arguments deal less with personal desire, and more with society.
"We spend 400 billion hours driving each year. We only use our cars 4% of the time," remarked Aicha Evans, formerly Intel's Chief Strategy Officer, now CEO of autonomous vehicle technology maker Zoox, during last June's Amazon re:MARS 2019 conference. "One-third of the traffic in cities is actually people searching for parking. One-fourth of air pollution is caused by transportation. Car crashes take the lives of 1.4 million people each year. Over 90% of those crashes are caused by human error."
In mid-April, Zoox released a video depicting how its autonomous vehicle approached a completely autonomous ride through downtown San Francisco, dodging pedestrians, double-parked vehicles, and suddenly opening car doors.
The world that supports and facilitates the scenario that Zoox's Evans foresees doesn't just substitute some post-middle-age Andretti wannabe with artificial neurons. This isn't just about an upgrade to people's Teslas. There are several grand infrastructural overhauls that the world will require before a transportation system capable of intelligently reducing pollution and saving human lives can even be beta tested, let alone safely deployed. For human error to be eradicated in any system, humans must first come to grips with the full extent of how humans have erred through the centuries -- and that won't be done without, of course, human error.
But if we take the first steps in that direction -- if putting an algorithm in everyone's driver's seat truly does inspire us to rebuild the world beneath us -- then perhaps the world would benefit from the changes it would entail, even if the full autonomous traffic system never comes into full fruition.
Moving targets
It was Netflix that revolutionized the programming model called microservices -- the capability to decompose a broader application into small functions, replicate as many of those functions as necessary, and run all of those functions asynchronously. But, at least for now, the Netflix model runs entirely in the Amazon AWS cloud. Netflix is distributed computing, if you accept the premise that the universe in which these services are distributed, is limited to one vendor's infrastructure.
A truly distributed computing system capable of operating a moving vehicle powered by an explosive substance can rightly be considered a mindless driver wielding a bomb. To the extent that it can make poor decisions, it is a deadly weapon. Yet the likelier event, given the state of our computing and communications technologies at present, is not so much a bad decision that ends up hurting someone, as a good decision that doesn't get executed in time on account of some unresolved bottleneck.
Part of the purpose of distributed computing is to make optimum use of processing power and network bandwidth, so that a very challenging task -- like watching the road and driving forward -- becomes manageable and efficient.
The value proposition for data centers for "edge computing" has been designing distributed workloads throughout a network so that processing takes place as close to the end user (the point of service consumption) as possible. This way, latency is minimized. This seems like a very sensible proposition, if that user makes sure to hold very still.
Machine learning systems are the most data-intensive workloads one can have in a network. When the data can be gathered through one access point, it's important that it too hold very still. Making the autonomous vehicle entirely self-contained is one option, but many would say it's a bad one. It stakes the life and well-being of the vehicle's occupants against the efficiency of a non-redundant control system. A safer alternative would work like a cellular communications network, which would delegate control of the vehicle to whichever remote server is closest to it at any given time. This would be an edge computing model at its most extreme circumstance, where critical tasks are "live-migrated," to use VMware's old term for it, through a long sequence of servers every few minutes, or perhaps even seconds.
But if the user is traveling down a highway at 70 mph -- and more importantly, is not the driver -- then the edge computing model would force the processor to relocate itself to the closest access point to the user for any given moment. It would appear the workload would be live-migrating more often than it would be processing, which is not a good situation to be in when a deer walks into the car's path.
The alternative approach would seem to be to keep the workload as stable as possible, perhaps only relocating it when latency levels are untenable. But that approach would assume latency is a one-way affair. Presumably the vehicle is transmitting a concurrent stream back into the network, via a gateway whose location truly does get relocated, like the access point in a cellular network. Some models depict the vehicle transmitting a complete, raw video stream. Others would have the car drawing inferences for itself from an on-board processor, which would stream a digested form of its interpreted world to the network, presumably at a lower bandwidth than raw video.
This is where we realize how early of a conceptual stage we're truly in, with respect to autonomous vehicle networks. The architects of 5G Wireless networks are working out the problem, as part of the 3GPP group's Vehicle-to-Everything (V2X) portfolio. But as Prof. Heath told ZDNet, communications architects and platform designers could stand to benefit from a little more input from the automotive industry.
"If we had a good 5G network now, the people who were driving Ubers could, in principle, be driving remotely," remarked Heath, "just like soldiers fly drones all over the world. It doesn't even have to be fully automated; it just has to be remotely connected. But we don't have that capability, and I think that capability was also overlooked."
The way communications standards are negotiated today are by consensus, of course, but specifically the consensus of its direct stakeholders. Heath asserts that the applicability of such standards could be improved if the industries with a stake in the applications -- most importantly, he believes, healthcare and automotive -- had more of a voice.
"In some cases like vehicular," he said, "there were organizations like 3GPP and the 5G Automotive Alliance, all coming up with requirements. But when I talk to auto companies, they say, 'Well, these requirements have nothing to do with what we want to do.' That's part of the problem." He conceded that automakers did attend consortiums' meetings, especially 3GPP's, but may never have been engaged in discussions on topics such as platooning. This is where a convoy of vehicles, such as freight trucks, share information and decision-making with one another, traveling along highways as a team -- one truck being the redundant back-up for the other.
It is a safe bet that the evaluation of whether to swerve to avoid a possibly animate object, and potentially endanger a variable number of other animate beings as passengers, should never be entrusted to a single processor. There may not be time, in an emergency circumstance, to execute a failover procedure. Platooning might resolve that issue, at least insofar as freight transportation is concerned. Multiple trucks with multiple processors could coalesce on who should take the lead and how it should handle a busy highway, much like team time trial cyclists.
Here is where a real-world application of autonomous driving starts to look more like Dr. Robinson's suggestion: a kind of auto-piloting supply chain management scheme.
"I totally think that it's going to happen," remarked Vijay Janapa Reddi, associate professor at the John A. Paulson School of Engineering at Harvard University, referring to the prospect of autonomous vehicles entering society during our lifetimes. "That doesn't necessarily mean it's going to take over every single thing that we know."
Prof. Janapa Reddi foresees progress with autonomous delivery vehicles for nationwide transport. He didn't name names, though the system he described sounded quite a bit like Embark, which logged nearly an eighth of a million miles in 2018 in autonomous truck testing, with evidence (unconfirmed) of Amazon's participation first reported by Commercial Carrier Journal.
"I think what these algorithms we've discovered are good at is being able to crunch massive amounts of data through," he told ZDNet in July 2019, "and being able to find patterns. That's a place where we [humans] struggle. We can only handle so much information being thrown at us, at a given point in time. But we have our own sets of perks. We can make these critical judgment calls, which computationally you may not be able to express. There's a balance there, but at the end of the day, the machine is still not very intelligent -- it doesn't really understand what it's doing. We understand things, but our responses tend to be slower, in a way, especially when we have large volumes of data. This human/computer interaction is going to be an interesting space that we should all be aware of."
Autonomous thinking
I have been, and may yet remain, an autonomous vehicle naysayer. The more necessary overhaul of our transportation system involves the implementation of a new power source, and an infrastructure for delivering it conveniently and practically. An optimum world of self-driving vehicles would be comprised entirely of autonomous driving, where decisions about itinerary, direction, and acceleration would be completely shared between mobile units. Speeds would be perfectly regulated, and conceivably stoplights at intersections would become moot, because vehicles would be pre-programmed to avoid one another.
Featured
Yet the safest, most cost efficient, most practical implementation of such a system would be busses and trains, not cars. And if we're going that route anyway, where's the real savings in replacing a few train engineers or superconductive shuttle pilots with an autonomous system?
I undoubtedly sound like the naysayers for the manned space program. I've spent my entire life, probably from the time I firmly recorded in my brain that W falls after V, defending the need for human beings in space missions. The naysayers talk about practicality, cost savings, logistics. And they make some sense. My rebuttal has always been that, once our species has decided to make the effort to work in space, we come together as teams and as societies to advance our science and our technology far beyond anything we'd imagined, with the benefits shared amongst all of us. "Software" is a concept that emerged from NASA, and so is cloud computing. Applying human ingenuity to resolving the issues of manned space flight led to the very creation of software. Making flight control centers more flexible and manageable led to servers in shipping containers, and the first scalable clouds.
Applying human ingenuity to the problem of vehicle autonomy -- of taking people out of the loop -- is leading us to re-engineer the problem of managing a transportation system in an entirely new and more approachable manner. Remote piloting, platooning, dynamic edge computing -- these are all benefits of working the science of autonomous vehicles, even if they don't end up leading directly to a self-driving car network.
In a near-term future where human contacts will need greater protection, and perhaps more substantive auditing and control, the driverless car could end up giving people greater control than they had before. In which case, I could turn my attention to some other pie-in-the-sky ideal, for new and more innovative ways to shut myself up about how pointless they are.
Stills taken from a one-reel movie produced by General Motors in 1956, entitled, "Key to the Future," now in the public domain, courtesy of the Internet Archive. Suggested by Brian Roemmele (Twitter: @BrianRoemmele).
Learn more: From the CBS Interactive Network
- Guide to autonomous vehicles: What business leaders need to know by Greg Nichols, Robotics
- For self-driving cars, winter is coming by Daphne Leprince-Ringuet
- Autonomous vehicles need well-marked streets more than 5G by Veronica Combs
Elsewhere
- Design Considerations for Autonomous Vehicles by Deepshikha Shukla, ElectronicsForU.com