Brad Templeton Home
When can this happen? (And how much?)
The most uncertain question is just when robocars, and the various other stepping stones in my timeline can happen. The technological uncertainties are complex, and the social ones depend on big variables as well.
Moore's "law," as interpreted by modern chip designers, should keep doubling information processing power every 18 months or so until at least 2020. This means a 256 fold increase in power by then over the best chips of 2008. More striking is the fact that the last factor of 100 comes in the final year of that 12 year run.
Robocars are mostly about computing, so robocars help bring Moore's law to cars. There was an ad long ago from a computer company that showed a picture of the moon and said, "If cars had advanced like computers, this would be your office view." Robocars can bring a bit of metaphorical truth to this.
Moore's law is a great helper of machine vision systems, but algorithms must be no slouches either. Improvements in this area are much harder to predict, though at this juncture each week seems to bring some impressive new work.
Yet none of this has had really big money behind it. The Grand Challenge teams have been dedicated, and talented, but the level of effort is still tiny compared to many of the other great research and engineering projects of history. Yet the payoff here is greater than that of Apollo in 8 years, and arguably even greater than the swift end to WWII brought by the Manhattan Project in under 4.
If the political community can become seized of this idea, it might be possible to make this a new "Apollo" project. This means a major political leader, such as the new President of the USA standing up and declaring, like Kennedy, that the nation to set itself the goal of ending car accidents before the 2010s are out. To do it, not because it is easy, but because it is hard.
Such a presidential decree would bring something far more important than funding. It would bring the political will, and it would bring the national debate and publicity required to win the hearts and minds of lawmakers and the public.
In the end, the public must accept robots on the roads with them, and then riding in those robots. And the public must accept the inevitable mistakes that happen along the way. The public's acceptance of the goals will be necessary for the law to change.
With such a bold public pronouncement, I believe 2020 is a possible time.
But there are some other routes, mostly outside the USA. Indeed, any major technological country could dedicate itself to this challenge. Not only would top developers from the USA and other different countries become interested in the foreign project, they could still work on it without leaving their hometowns in the modern networked world.
However, significant advances in other countries could quickly make the USA and other countries become afraid of missing the boat (or rather the robocar) on an important new life-saving and green technology. Early deployment in India or China would make the USA very jealous. This would hasten the public's willingness to adopt the technology. These other countries, for better or for worse, might be more willing to politically tolerate the small number of people killed or injured due to engineering errors in the early phase. Once the problem rates get very low in those countries, adoption in the industrialized world becomes easier.
Another thing that could speed up development would be pledges to buy large numbers of safe robocars. The most likely scenario would be the military promising to buy large numbers of robocars, deliverbots or associated technologies. However, goverments could also promise to purchase large numbers of robocars to do things like move federal employees around Washington or other capitals.
Cities could also purchase fleets of robocars as an alternative to buying expensive transportation systems. For example, instead of paying a billion for a light-rail line, a city could deliver better service in that corridor by purchasing a few thousand robocars for far less money.
Work is in progress
Many of the world's major car manufacturers have autonomous vehicle labs. Most notable are Volkswagon, Volvo and Nissan. The Volkswagon labs has a project modifying an Audi TT so that it can try to win the Pike's Peak Hill Climb challenge -- a timed trip to the top of Pike's Peak.
Google has also built is own sophisticated robocar lab, using some of the most experienced developers. In October of 2010, they announced that their modified Prius cars had performed 140,000 miles of supervised driving with minimal supervisor overrides, and gone for stretches of over 1,000 miles on ordinary highways and streets with traffic with no human intervention.
I've outlined the concept of a whistlecar -- a car that brings itself to you when you summon it, typically via your cell phone. It could be your car or one you are hiring from a taxi company. It drives itself to your location, but then you get in and drive it yourself. It almost certainly has lots of accident avoidance technology so you don't have many accidents while in it, but human error remains possible.
On one hand, a whistlecar seems an easier thing to build. While moving itself, it does not need to go very quickly, and can limit itself to a subset of the roads. If it can't go on the summoner's street, it may be able to get near enough to be a quick walk -- something more akin to a transit line. In fact, whistlecars might very well use existing transit lines or even follow close behind buses and streetcars (or even other cars) to get to their destinations. They might be equipped with special bumpers or even external airbags to make them safer.
With no humans inside, whistlecars could stop or swerve very suddenly so it's easier to make them safer to put on the road, as far as hitting people is concerned. Keeping to slow speeds and dedicated lanes, and only following other cars through intersections and crosswalks, they could eliminate the risk of hitting pedestrians. If these early vehicles were to hit other cars or inanimate objects, there would be damage, but it's something insurance can solve as long as designers keep the rate of this low.
If whistlecars can be ready first, the question is whether they can be accepted first. Whistlecars can bring us many of the benefits of robocars. They can provide the right vehicle for a trip, including a single person electric vehicle. They can avoid the hassles of parking and refueling. And while not perfect, they should be among the lowest accident cars on the road even while humans drive them.
However, they miss out on a number of the advantages, such as letting the non-passenger work or read, full avoidance of human error and superior packing to reduce congestion. In fact, they can increase congestion, because every self-delivering whistlecar, if not replacing a human-driven taxi trip, is going to be an extra car moving on the road.
They do however really boost the convenience of carsharing. Car sharing is hourly car rental from hopefully nearby local depots. Carshare cars that come to you would be fabulous, and one way trips could also be enabled where today most carshares require returning to the same lot. This means you have fewer cars per person in a city, which turns out to have a lot of green consequences, since about 10-15% of the lifetime energy cost of today's cars is spent in their manufacture.
An unanswered question is how the public would deal with vacant cars roaming the streets, moving to customers. Like a fleet of taxis with no drivers, but smaller and greener. Unfortunately, any problem with the whistlecar software systems that injures or kills a person comes without the vastly greater saving of human life that removing human drivers brings to compensate. The whistlecars will save lives, but so can a non-self-delivering regular car with the same accident avoidance technologies.
If the whistlecars are allowed on the roads (and not just on transit tracks) then their safe operation will prepare people for the days of robocars more quickly.
Robocars will probably face problems not yet conceived of, but even if not, many scenarios are possible where they are delayed for decades. If the technology becomes workable, it does seem that they should appear somewhere in the world sooner, even if only in rural areas or military zones.
However, there are a few nightmare scenarious which could scare the public away for a very long time, such as a computer intrusion attack that manages to hurt large numbers of people, or some cascading failure which causes an entire drafting train on the highway to crash and kill people. Regular use as terrorist weapons could see them banned.
Short of these problems, the march of Moore's law and improvements in machine vision and AI seems inexorable. These technologies, and robotics in general, are too valuable, too useful for so many other aspects of life that they will not be stopped. Even if one year machine vision and comprehension seems impossibly far away, this could suddenly change in a very short time due to other driving factors.
Today the technology used in prototype robocars is expensive. But last year's $2,000 computer is this year's $1,000 model. All the computer technology will eventually become a minor part of the price of a serious automobile. Anything electronic made in quantity one billion becomes very, very cheap.
Small electric cars, however, are very cheap to make and maintain, if you don't consider the batteries. They need far fewer parts, and far fewer moving parts. Drivetrains are simpler -- there may not even be a transmission. Everything is drive-by-wire with computer controls.
Some models of robocars will be cheaper because they don't need things found in today's cars. An urban car that never goes on the highway and rarely hits red lights doesn't need to accelerate quickly, or go at high speed.
Once the roads are safe, cars need not be heavy for safety reasons. They can be made of light, composite materials instead of energy-intensive steel.
Deliverbots, with fewer safety constraints, will be even cheaper. They may be not much more than a frame, cover and motors. They don't care at all about acceleration or creature comforts.
The energy costs of travel in light electric vehicles are well below 1 cent per mile. If a vehicle has a service life of 200,000 miles, and can be built for $15,000, that's 7.5 cents/mile (with no discount rate factored on a short-lifetime taxi.) This suggests a travel cost well under 10 cents/mile, or well sub-dollar for urban trips and deliveries.
In fact, I suspect the vehicles and deliverbots can be made for much less than $15,000. Their lifetimes may also be longer, though this is harder to predict. Even with taxi company margins, urban trips will be very cheap compared to today.
Now consider consequences of the deliverbot.