Wednesday, August 31, 2016

Early Indications August 2016: The Next Car

About 125 years ago, when the internal combustion engine supplanted equine power for personal mobility, there was much talk regarding “horseless carriages,” defining the future in terms of the past. We are at much the same juncture today: as electric autonomous vehicles come closer and closer to mass-market availability, much of the conversation starts with what we know human drivers do: “How will self-driving cars avoid bicyclists? How will self-driving cars merge in construction zones? How will self-driving cars make left turns across oncoming traffic with solar glare?” All of these questions must be answered, of course, but I believe it’s not too early to ask what we want of the _next_ car, the one(s) with a largely new set of constraints and capabilities. That is, given a clean-sheet design, what are some questions we might ask? Here are three among many.

1) How do we balance autonomy with “mesh transportation”?

By definition, a driver in a car is largely autonomous and disconnected from the cognition of those around; “what was he thinking?” is a common complaint while observing other drivers. The person at the wheel can follow or ignore traffic laws, brake suddenly or gradually, act with awareness of other vehicles or possess limited situational awareness. There are many consequences of this autonomy: cars have long been associated with personal “freedom,” traffic flows in an annoying and predictable accordion pattern in congested stretches of highway, and of course accidents happen when driver A somehow surprises driver B.

Once driverless vehicles constitute some critical mass of traffic, however, that assumption of autonomy can be challenged. My current frame of reference is a mesh wireless network, a potentially peer-to-peer ad hoc configuration of cars both interacting with vehicles close to them and serving as repeaters for less proximate “nodes.” A simple scenario started my thinking: what if a truck stuck in traffic wanted to see the sensor feed from the car at the front of the pack? Already, Samsung has shown a heavy truck with an LCD display on the rear showing the view out the windshield. Once my vehicle can “see” the sensors n cars ahead, what else can happen?

Although sensor-driven autonomous vehicles are getting substantial attention from Uber, Google, and Tesla, the notions of vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) are also gaining mindshare. Already emergency vehicles can trigger stoplights at some intersections (green for the ambulance, red for the cross-streets), it doesn’t require an enormous leap of imagination to get to cars addressing and listening to the signaling infrastructure: why train a self-driving car to understand (or memorize) speed-limit signs when the information could be available via wireless beacons? The appeal of V2V is obvious, as per the “what was he thinking?“ scenario above: if vehicle A can signal intent with more lead time, more consistently, to more of the relevant peers and bystanders, the safer matters should become.

For years, European companies have been trialing road trains: a lead heavy truck invites close followers (who hand control off to a wi-fi network) to ride in its wake. The tighter following distances improve fuel economy and free drivers to attend to less mundane chores. Now what happens when self-driving vehicles can self-assemble? If 8 or 10 cars and trucks all enter I-95 from around Princeton going northbound, what if they form a road train until the first of the vehicles comes to its exit? And what if there’s a mechanical connection, like those magnets on wooden train sets? Road utilization improves, fuel economy improves, and the lack of human drivers means nobody is bothered by the unappealing view of the vehicle ahead.

Given the currently lousy security of Internet of Things things in general, and of wireless car systems in particular, for this mesh-vehicle future to be safe, there will need to be massive strides in, and probably complete rethinking of, security practices. This means clearly understanding the trade-offs that are being asked and granted. The security domain thus echoes the larger debate that will emerge as to what vehicles can and cannot, should and should not, tell and learn in interaction with other vehicles and the world.

2) What happens when cars don’t project personal identity?

I know something of my readership. You’ve driven stealth M3s, E-types, Z06 Vettes, and of course 911s. This is not mass-market car-dom, but the extreme proves my point. In many cases, vehicles are designed as much as a projection of the owner’s psyche as they are for road performance. Compare a Suburban to a minivan, or a new pickup truck to a 30-year-old version. Absent towing capacity, the functional performance may be similar but the features, price (I didn’t say cost: GM builds massive profit into that Suburban), and enhancement of one’s personal brand are very different.

Auto executives have already begun publicly worrying about vehicles designed and sold only as appliances: if people buy transportation as a service rather than as a product, the design remit changes. As a rider, I certainly prefer the S-class Mercedes taxis common in Europe to the cramped Priuses (Prii?) I get in US cities. Do I insist on a certain type of car to haul me around, especially when nobody will see it in my driveway either way? No way. Once autonomous vehicles are optimized for whatever we decide to optimize (please make the London taxi one of the blueprints . . .), conspicuous consumption will fall far down the list, particularly for mass-market cars and trucks. (Read more here and here.)

3) What happens if safety is reset to a higher priority?

There’s a famous assignment given to engineering students: design a protective enclosure such that the egg inside it can survive a fall of a specified height. I thought of it immediately when I saw this effort by an MIT team to understand safety trade-offs to be encoded in autonomous vehicle algorithms.

Why the egg drop? All of the Moral Machine scenarios embed numerous assumptions; I wanted to challenge those, not take them as given. The passenger cocoon could be one such: when we design, license, and support different vehicle designs, what do we want optimized?  Should pedestrian safety be a higher priority than what happens to a car’s passengers? Why or why not? To give an example, Google has a patent on a flypaper-like technology to snag people on car hoods after they’ve been struck. How will we set the weights of protection for passengers, cargo, pedestrians, bicyclists, driveway shrubbery, and other features of the driving environment? How much will this be done by markets, how much self-policing will we see, and how much government regulation will be imposed?

If safety is a higher priority than it is today, why have windows at all? Boeing has suggested replacing airplane windows with display screens, so why not equip cars the same way? At the same time, if passengers are cosseted in metal tank-like vehicles, what unintended consequences could there be for driving algorithms, bystanders, and even fast-food restaurants?

Just as interstate highways, McDonald's, and suburban sprawl could not have been foreseen by Daimler, Ford, and Durant, we will see parallel discontinuities in the coming century. Railroad ownership, to take one example, went from being the source of massive wealth and prestige (as with Stanford or Vanderbilt) to a joke (Conrail) in a relative short time. Entirely new ancillary services and industries emerged, and may now recede: will anyone become a billionaire owning parking lots in the next 30 years?

A key question relates, as it always does, to the speed of the transition. From an engineering standpoint, designing a world of only autonomous vehicles would be relatively easy; similarly, we know how to build roads, vehicles, and venues for cars with people driving them. The mixed zone, to borrow a term from the Olympics, is what we currently face, and it’s one hard problem on top of another:

What are the business models? (Google is struggling with this as I write, having replaced a respected-but-departed roboticist with an AirBnB executive to run the car project.) Who bears liability? Who pays to upgrade the infrastructure, whether to fill potholes or install beacons in road signs? What signaling conventions can carry over and what new ones need to be designed; can the camera of a self-driving truck reliably see a turn signal or is a radio message more appropriate? Where will trusted suppliers come from (Panasonic, a partner in Tesla’s battery factory, is emerging as a big player in the global automotive ecosystem, for example)? What about car loan companies, service bays, and other businesses whose mission will be redefined? How will gas stations as real estate, and oil companies as businesses, be forced to evolve (charging a battery takes a lot longer than filling a tank: how many 20-pump gas stations will become 20-plug chargers? Can local electric grids handle such concentration?) Which brands will win and lose? Why? How will human-driven cars fade into antiquity? Will valets, for example, be kept on for ceremonial value at high-end destinations? Which internal-combustion vehicles will have the longest carriers in the new era, and which will be most prized by collectors? What happens to the freed-up parking and other infrastructure? Does congestion pricing a) disappear or b) enter a new phase of complexity? What happens to municipal revenues dependent on traffic tickets? How will domestic architecture evolve without the need for the same size and type of garages? Where will housing be located relative to work? The list goes on, but suffice it to say, Detroit and its extended network is in for the shock of a lifetime.

Postscript: As I was wrapping up this newsletter, tech analyst Benedict Evans of Andreesen Horowitz posted a series of Tweets asking many of these same questions.