Brad Templeton Home
Objections and Answers (FAQ)
Are you crazy? This is totally Science Fiction
Obviously this technology is not yet delivered, and predictions about future technology are risky. I recommend you watch the videos about the current prototype robocars that are winning the military contests, look at Google's robocars and read other materials to decide if and when this is doable.
This is not an easy problem. It's a hard problem. But more and more evidence suggest it is a hard but tractable problem, given the right resources and political/social will.
This is full human artificial intelligence, it's many decades away
Driving does not require human level intelligence. In fact, swarms of insects can navigate among their own kind with very little vision or brain power. Humans can, and do, drive in their sleep. Dogs, horses etc. can navigate fine in crowds of people. Robocars don't see like people, think like people or drive like people.
Again, this is not an easy problem, but nor is it a far-future one. Truth is, Darpa grand challenge teams already have protype vehicles that can do this, though imperfectly. Today.
You should take care in your thinking to avoid linear extrapolation, imagining that things improve at the same pace all the time. In fact, with computer technologies, the rate of improvement itself gets faster every year. That means that in a 10 year plan, half of the improvement can happen in year 10, and those who made predictions based on how things were going in year one were way off.
Americans love to drive cars
Certainly a large class of Americans do love their cars and love driving them. But in many large cities of the world, including New York City, most people don't even own cars, and a large portion don't even have drivers' licences. Indeed, New York and many other towns are already partly into the robocar world, in the sense that the middle and upper classes travel almost exclusively by hailing cabs.
So it's clearly possible that people can be convinced to embrace a world where they don't drive. However, for several decades, humans who want to will still be driving, and they may do this forever on fun and windy mountain roads. However, there will be a limit to how much we will tolerate a love of driving and the large number of deaths that come from it when there is another choice.
Are you crazy? My computer crashes all the time!
Indeed, making reliable driving software is one of the central challenges of the robocar problem, and I talk about it in the roadblocks chapter. However, you should not conclude that just because a personal computer OS is unreliable that it's not possible to build embedded robot software that is. In fact, any modern car contains a fair number of computers doing all sorts of things necessary to make it run, and there are lots of embedded computers running life-dependant systems in airplanes, hospitals and many other places.
The problem is hard, and takes time and money and research, but it is not that similar to the problem of building Microsoft Windows.
GPS and road data aren't reliable
The Robocars I envision will certainly make use of GPS and road data, just as human drivers make use of maps. But it will be just one of many inputs they use to drive and navigate -- and mostly to navigate. Robocars will be making use of their own self-contained sensors such as many cameras, LIDAR, sound, radio, intertial motion, slope and anything else designers can think of to track their environment. They'll see lane markers, traffic signs, traffic lights, street names and much else just as drivers do. They'll be in communication with other cars.
They will also use GPS, but they won't use just the one decades old system we have now. Soon there will be more than one GPS system in the sky, and newer ones will be more accurate and more secure. There will also be local transmitters in cities providing extra location data as well as super-accurate data that lets them get position to the centimeter when it's available.
All those factors will be combined to make decisions about driving (how to move on the road) and navigation (when to turn onto other roads.) The failure, or even compromise of a few of those sytems will not prevent safe operation -- or if it does, the vehicle will stop safely to allow human navigation.
Road data can get more reliable very quickly if every car that notices a difference between the real road and the road data reports it in. When there's no road data (ie. dirt roads or private drives) a human could still navigate (with a wheel or joystick) and thus build it.
Google's current robocar project in fact is based on using tools combining their existing road data projects and detailed robocar sensor collection of enhanced road data. They drive every road first with a human driver and get complete maps of where everything is -- lane boundaries, curbs, trees, buildings. This lets them figure out where they are and what to do with great accuracy, even without GPS.
People keep stuff in their cars, they want to own them
I'm certainly one of those people. And many people will own a car for exclusive use. However, again examples like New York City show that even in the USA, people can quickly adapt to not owning cars.
As an alternative, I have proposed the idea of a standardized "stuff locker" which fits in a robocar and can be quickly transferred from any car to another car or to a robotized locker room. Not quite as good, and you have to be less sloppy, but it presents at least part of an alternative.
Nobody will own a car when you can get a robot taxi on demand
Certainly the ability to summon a robot taxi quickly with a mobile phone will make many people decide not to bother owning a car. It will be much cheaper not to own. But for various reasons, I suspect many people will keep on buying, because they are wealthy enough to afford it. Or they may buy a very nice car and hire it out when not using it.
People view a car as an extension of themselves
Once again, clearly there are many places where people get over that quickly. But robocars actually can offer more ways to allow people to express themselves in their transportation, if we develop things like car clubs, subscription car services and cars that customize graphics and more for their occupant. It's not impossible to imagine that electronic ink might allow the occupant to command the taxi, like a chameleon, to switch to a custom paint job just for them.
We should redesign our cities to be more efficient, not get better cars
It's true, our 20th century cities, which grew up around the car, are not designed for a world of efficient transportation. And we should work on improving that. While new development can be better designed, reworking what exists will take many decades, even centuries. It has to be done with politics, and urban planning, and decade long time horizons and multi-billion dollar budgets.
So we should certainly think about that -- though I predict that as robocars arrive, our thinking will have to be re-thought very frequently. We need solutions that can work in the cities we have now, with the people and desires that exist now. The truth is that efforts to "design" cities usually end up with results quite different from what the designers envisioned.
We need more transit, not more cars
Transit can be efficient, and is in Asia and Europe, but even there it has a hard time beating out ultralight electric vehicles. In the USA there's not even a contest.
But the real contest is between two ways of thinking: Central planning and grass-roots revolution. When it comes to computer technology based change, nobody even debates which one wins any more.
City redesign and transit procurement takes decades. It is done by politicians, and full of politics. The time horizons are huge. The budgets are huge. By the time the projects are completed, the rules have changed.
When ordinary private citizens can go buy technology at the store, they just buy what they like. They have no plan. They react instantly. And computer technology gets twice as good every year or two, which makes it good that they aren't planning. The revolution comes from the ground up. It seems wasteful, as people buy things that are obsolete a couple of years later.
One great example of this is 802.11 (Wifi) which uses unlicenced spectrum around the band that leaky microwave ovens broadcast. Crippled as that spectrum was, people just built and bought products, and it resulted in the greatest revolution in the history of radio. Robocars -- once legal -- offer the chance to apply that methodology to transportation. Just as computers of today are unrecognizable compared to those of the past, cars can find the same fate. I don't know what that fate is, I just know it will be far more interesting than anything that comes from central planning, where the systems are obsolete before they break ground.
Won't robocars need central control?
Many visions of self-driving cars have imagined a central computer controlling and coordinating the activities of the cars.
It is more likely that the cars will be autonomous, able to drive without any communication with a central system or other cars on the road. This is of course what humans do when we drive, what animals do when they school or swarm. In fact, this is almost surely a "must" since the first robocar is not going to have a central control system made for it.
This does not mean that such communication won't happen, or be highly useful, just that it won't be necessary. When the first robocars hit the roads, there will not be any central traffic control computer even if we wanted to build one, and the other cars would not communicate.
Robocars will of course receive information from traffic signals, and even more advanced information over computer networks about traffic patterns and the planned timings of traffic lights. (Humans have always gotten traffic updates over the radio too.) However, like the humans, they will not depend on being able to send information back.
Now there are some great uses for 2-way communication and this probably will happen, again without being necessary for operation. At the simplest, cars can report back information on traffic condition and traffic hazards, and other cars can learn from that. They could report road conditions, with exact positions of icy patches or potholes.
And yes, they could announce their intentions to the other cars around them -- as we do with turn signals and brake lights. They will do those lights of course, but there is no harm in backing it up with another channel.
Finally, they could start trusting messages from other cars to reduce congestion. A robocar, driving the road near an unknown (probably human driven) car would leave enough room so that it could get out of the way if the unknown vehicle did something erratic, like swerving or sudden braking. Because the robocar's reaction time is so much better than a human's, it could get away with a very small gap in many cases.
However, it would have to also have to know that if it had to swerve to avoid something, there would be a gap into which it could swerve. If it were able to enter into a trust relationship with other robocars on the road of the form, "If I have to swerve to the right, into your path, can you also swerve with me?"
The vehicles could model how they will react to anything unusual -- unknown cars, pedestrians entering the street etc. -- and always know they have an escape plan. Ideally, and always at first, they will have an escape plan which does not involve trusting much about the other cars. However, as systems get more reliable, being able to trust those cars would allow more densely packed roads, increasing road capacity.
We have to switch to robocars all at once
There's no way that we will be able to switch overnight from human driving to robot driving. The first robot car on the roads will need to deal with every other car being human driven, and the potential for pedestrians to use crosswalks and to jaywalk.
Robocars must be able to do all that a human driver does, and in fact largely do it much better. They must be able to see in 3-D (though humans can still drive with one eye, though not quite as well.) They must track all other cars, people and obstacles on the road. They must be able to spot and decode lane markers, traffic lights and all road signs, even in bad visibility.
And, in fact, they are already on the way to doing this. (Though the current cars tend to cheat by just knowing in advance where all the road signs, traffic lights and such are.) However, today's robocars can and do track and identify all other cars, pedestrians other moving objects.
Robocars will need new highways and special roads
Some robocar plans, including a test track in San Diego, and the French and European Cybercar initiatives involve building specially modified lanes for robocars. For example, one could bury a wire down the middle of the lane and make it easy for a robot to follow the wire. (This has been done to control transit buses where the driver does not need to steer, just use the gas and brakes. As a plus, the buses always stop perfectly aligned with the curb making disabled entry much easier.) And of course, the first commercial semi-robocars from ULTra run only on a special isolated track.
Some propose special markers on the sides of lanes that are easy for machine vision to find. One could also put special codes on all road signs. (Mile markers would be handy to let vehicles get much more exact positions on the road compared to what GPS does.)
All of these would be good ideas, but they are expensive and would not be ubiquitous at the start. And truth be known, the technology to follow existing lane markers and road data is already very good and should be 100% solid before the other technologies are ready. One of the big advantages of robocars is that they provide quality green transportation without having to put in publicly funded and approved infrastructure.
This is not to say that we won't see test projects using special lanes, and we already have. They will provide useful information and will serve their limited areas. I believe, however, that they are a distraction. Special lane robocars can't readily deliver the robocar vision I outline because they don't go everywhere. Driving still becomes necessary to go everywhere, which makes the taxi/carshare concept unworkable. (You can drive a car from the special lanes to your house, but it is unable to then move to the next customer; the next customer has to be very nearby and able to come to it.) Unfortunately a special-lane robocar is only valuable to you if you travel frequently on those routes. It is far less marketable than an anywhere-to-anywhere robocar.
It's also much cheaper to just put marks in a database than it is to record them in the physical world.
The real value of special lanes would come if human drivers (and pedestrians) were banned from them. The absence of human drivers in the lanes would allow a more predictable world for the robocars, and less fear from the public, and perhaps fewer liability issues. So the reality it is human drivers who will need special lanes.
Will we have highway autopilot first or city driving first?
Some predict that we'll see robocars first on highways, and then the drivers will need to take over for the city driving. There is no question that highway driving is a much simpler task than city driving, so there is merit to this argument. Making special highway lanes is also more practical than inserting special markers into major city streets. (Indeed, you can make a track a robot can follow quite easily with a machine that just piledrives a small magnet into the pavement every 5 feet.)
Highways also don't have pedestrians or cyclists to worry about.
However, highway driving is much faster, and thus mistakes are far more dangerous to vehicle occupants. In particular, the stopping distance is much larger -- 300 to 400 feet, compared to under 100' on city streets.
This turns out to make a big difference with present sensor technology. LIDAR, which is a laser radar. LIDAR is very reliable at detecting possible obstacles, but is only really good at that today to about 130 feet out. Highway driving requires detecting potential hazards and issues much further away, which must be done with other technologies, typically machine vision from cameras. Vision is of course what humans use, but today, machine vision will not detect everything.
So at 30mph you can build a robot car that will reliably detect something unusual in its path, and can stop in time, but that's a lot harder at highway speed. These technologies are evolving, however, and I expect improvements to come in both that will change this equation. However, it does mean that urban cars and deliverbots might be safe sooner than highway cars, in spite of how much easier highway driving is.
I do believe, however, that robocars will work better if we convert more roads to one-way streets. The high convenience of the robocar that drops you at your destination and zooms away will make the slight extra distances of one-way streets a minor issue.
With one-way streets, the robocar does not have to worry that the vehicle rushing towards it in the other direction might suddenly veer into the oncoming traffic. This is something even the robocar can't avoid if it happens. The robocar will assume, as all human drivers do, that the cars in the other direction will not cross the line at high speed. (They will make left turns and occasional slips, which will be prepared for.)
One-way streets solve the problem, and also help reduce congestion and fit more half-width or 2/3rds width cars on the road.
Electrified transportation is more efficient
Transportation that gets power from overhead wires or 3rd rails is more efficient because it doesn't need batteries. That saves the weight of the batteries and the losses in charging them. It also means that they can travel as long as they like (while on the power network.)
But all this comes at a cost too. One needs the power network of course. To go off the network -- which is vital, as I describe above -- you need a battery anyway, though a smaller one that you recharge while on the network. Or you could use multi-vehicle trips with a different vehicle for off-network travel.
The cost of the power network is high, and based on what we see in cities with overhead trolley lines, quite ugly. It doesn't have to be ugly but we seem to be willing to be ugly to save money. And network-only vehicles have limited mobility to move around obstacles and avoid things on the road.
Using rails on the ground is not ugly, but it's costly. Rails for drive are noisy, but you could have rails just for power. Today's 3rd rail technology is not safe to use on open streets, but I do believe that it could be designed to be safe, with tiny rail segments that only offer current when the vehicle is directly above them.
However, I think such approaches are better as an add-on. Once you have working electric vehicles with batteries, you can consider power lines to make them better. We already know how to make good, short-range battery vehicles. No revolution is needed there.
Without transit, the roads will clog completely
It is true that adding capacity to roads soon leads to more usage. But robocars add a lot of capacity to the roads.
Human driven cars take a full lane for an average of 1.5 passengers. Robocars with one passenger will usually be half the width of a lane. (Many robocars with 2 face-to-face passengers will also fit in half a lane.) Human driven cars require several car lengths of space between them. Robocars can follow one another at very short distances. With just those two changes you can send around 6 times the number of people per lane than you can with human driving.
To top it off, robocars should be able to go faster, which increases capacity. They won't get into congestion if they can avoid it, will always merge smoothly and will distribute over streets by tracking traffic volume. (When traffic collapses to stop-and-go, road capacity is cut severely. Robocars should be able to avoid that.)
If all this doesn't help, then we can have transit at rush hour, but much more pleasantly. Imagine 30 single person robocars converge at the same time at a bus stop next to the highway. Quickly the passengers get into the bus, and their cars go off to carry other people. When the bus gets to its destination, 30 different single person robocars are waiting for a quick transfer to the final destination. With this system, the roads can handle many times more people.
I have a more detailed article on congestion and capacity.
Won't the empty Robocar "car moves" be wasteful?
When robocars travel without an occupant to deliver themselves, or to refuel, this will indeed waste energy. It's hard to predict how much, but I think the amount will be reasonable. First of all, when we get to ultralight robocars, the car could weigh just a few hundred pounds, and thus will be more efficient when empty. We also make wasteful side-trips today to find parking or fuel.
At some point we will decide between having more vehicles (thus reducing the need for unoccupied trips to pick up people) or fewer vehicles and more car-moves. The market will find a balance. Light robocars will be so efficient that even if they spend half their time unoccupied we're still doing a lot better than transit or gasoline cars.
How well we do will depend on traffic patterns and how well we predict them. Good predictive software may be able to cut down the empty car moves a great deal. Of course, paused robocars will use minimal power, unlike idling taxis.
Another thing that makes car moves more expensive than having more cars is the extra range needed for car moves means bigger batteries, which are the most expensive thing in an electric car.
What about all the strange things that happen on the roads?
Most driving is pretty mundane, but strange things happen. Plastic blags blow across the road and it's OK to drive through them. Erratic drivers go nuts. Guys with spray bottles try to clean your windshield. Cops and civilians sometimes direct traffic by hand. Carjackers point a gun at you. And a thousand other things.
That's why it is not sufficient to just do mundane driving. But every odd scenerio anybody's heard of will be going into the simulator and test suite, with a few more you haven't heard of. In particular anything that's ever happened to a robocar and was recorded by it is going in as well, modelled on the real world data.
This means that every robocar software system has already driven through, in the virtual world, every situation anybody has thought of or encountered. Driven though every variation of it that simulator authors could imagine too. And done the right thing, or it was fixed. With people, when one person learns something about a new situation, everybody else still doesn't know. With robots, they all will be informed quite quickly. Every robocar system will have driven every road you will ever take it on, in simulator, before you get in.
Still, there will be situations the car can't handle. If the situation involves something urgent, like potentially hitting something, the car must always be able to handle it, even if it has to handle it by pulling over and stopping if for some reason a human can't take over. We don't want the cars on the road unless they are able to do that when they can't figure things out, and we'll not want to buy them if this happens with any frequency.
For non-urgent problems, like a new road sign the car doesn't understand, or a detour, it's OK to ask a human -- either the occupant, or a remote human if there is a network connection that can be used to send video of the situation. And that human may advise that other people will also need to be consulted if other cars come to the same situation. Except now, the car will ask in advance, before it gets confused. Only the first car to see the new sign or detour or obstacle has to pull over and wait.
The Rent-on-demand deliverbot idea is crazy
I have indeed had many people feel that the idea of deliverbots bringing durable items just-in-time to your house is not workable. I still think it is an interesting idea, though tangential to the real message of these essays. I agree there is much to be resolved, but I do remain convinced that cheap deliverbot technology would change a lot about how we use goods and services, and how we live in cities and suburbs.
In the years since this first list of objections was written, more common myths arose, which led to part 2: Top 10 myths of robocars.