Submitted by brad on Thu, 2013-10-10 14:15.
More and more often in mainstream articles about robocars, I am seeing an expression of variations of the classic 1960s “Trolley Problem.” For example, this article on the Atlantic website is one of many. In the classical Trolley problem, you see a train hurtling down the track about to run over 5 people, and you can switch the train to another track where it will kill one person. There are a number of variations, meant to examine our views on the morality and ethics of letting people die vs. actively participating in their deaths, even deliberately killing them to save others.
Often this is mapped into the robocar world by considering a car which is forced to run over somebody, and has to choose who to run over. Choices suggested include deciding between:
- One person and two
- A child and an adult
- A person and a dog
- A person without right-of-way vs others who have it
- A deer vs. adding risk by swerving around it into the oncoming lane
- The occupant or owner of the car vs. a bystander on the street — ie. car drives itself off a cliff with you in it to save others.
- The destruction of an empty car vs. injury to a person who should not be on the road, but is.
I don’t want to pretend that this isn’t an morbidly fascinating moral area, and it will indeed affect the law, liability and public perception. And at some distant future point, programmers will evaluate these scenarios in their efforts. What I reject is the suggestion that this is anywhere high on the list of important issues and questions. I think it’s high on the list of questions that are interesting for philosophical class debate, but that’s not the same as reality.
In reality, such choices are extremely rare. How often have you had to make such a decision, or heard of somebody making one? Ideal handling of such situations is difficult to decide, but there are many other issues to decide as well.
Secondly, in the rare situations where a human encounters such a moral dilemma, that person does not sit there and have an inner philosophical dialogue on which is the most moral choice. Rather, they will go with a quick gut reaction, which is based on their character and their past thinking on such situations. Or it may not be that well based on them — it must be done quickly. A robot may be incapable of having a deep internal philosophical debate, and as such the robots will also make decisions based on their “gut,” which is to say the way they were programmed, well in advance of the event. A survey on robohub showed that even humans, given time to think about it, are deeply divided both on what a car should do and even how easy it is answer the question.
The morbid focus on the trolley problem creates, to some irony, a meta-trolley problem. If people (especially lawyers advising companies or lawmakers) start expressing the view that “we can’t deploy this technology until we have a satisfactory answer to this quandry” then they face the reality that if the technology is indeed life-saving, then people will die through their advised inaction who could have been saved, in order to be sure to save the right people in very rare, complex situations. Of course, the problem itself speaks mostly about the difference between “failure to save” and “overt action” to our views of the ethics of harm.
It turns out the problem has a simple answer which is highly likely to be the one taken. In almost every situation of this sort, the law already specifies who has the right of way, and who doesn’t. The vehicles will be programmed to follow the law, which means that when presented with a choice of hitting something in their right-of-way and hitting something else outside the right-of-way, the car will obey the law and stay in its right-of-way. The law says this, even if it’s 3 people jaywalking vs. one in the oncoming lane. If people don’t like the law, they should follow the process to change it. This sort of question is actually one of the rare ones where it makes sense for policymakers, not vendors to decide the answer.
I suspect companies will take very conservative decisions here, as advised by their lawyers, and they will mostly base things on the rules of the road. If there’s a risk of having to hit somebody who actually has the right-of-way, the teams will look for a solution to avoid that. They won’t go around a blind corner so fast they could hit a slow car or cyclist. (Humans go around blind corners too fast all the time, and usually get away with it.) They won’t swerve into oncoming lanes, even ones that appear to be empty, because society will heavily punish a car deliberately leaving its right-of-way if it ends up hurting somebody. If society wants a different result here, it will need to clarify the rules. The hard fact of the liability system is that a car facing 5 jaywalking pedestrians that swerves into the oncoming lane and hits a solo driver who was properly in her lane will face a huge liability for having left their lane, while if it hits the surprise jaywalkers, the liability is likely to be much less, or even zero, due to their personal responsibility. The programmers normally won’t be making that decision, the law already makes it. When they find cases where the law and precedent don’t offer any guidance, they will probably take the conservative decision, and also push for it to give that guidance. The situations will be so rare, however, that a reasonable judgement will be to not wait on getting an answer.
Real human driving does include a lot of breaking the law. There is speeding of course. There’s aggressively getting your share in merges, 4-way stops and 3-point turns. And a whole lot more. Over time, the law should evolve to deal with these questions, and make it possible for the cars to compete on an equivalent level with the humans.
Swerving is particularly troublesome as an answer, because the cars are not designed to drive on the sidewalk, shoulder or in the oncoming lane. Oh, they will have some effort put into that, but these “you should not be doing this” situations will not get anywhere near the care and testing that ordinary driving in your proper right-of-way will get. As such, while the vehicles will have very good confidence in detecting obstacles in the places they should go, they will not be nearly as sure about their perceptions of obstacles where they shouldn’t legally go. A car won’t be as good at identifying pedestrians on the sidewalk because it should normally never drive on the sidewalk. It will instead be very good at identifying pedestrians in crosswalks or on the road. Faced with the option to avoid something by swerving onto the sidewalk, programmers will have to consider that the car can’t be quite as confident it is safe to do this illegal move, even if the sidewalk is in fact perfectly clear to the human eye. (Humans are general purpose perception systems and can identify things on the sidewalk as readily as they can spot them on the road.)
It’s also asking a lot more to have the cars able to identify subtleties about pedestrians near the road. If you decide a child should be spared over an adult, you’re asking the car to be able to tell children from adults, children from dwarves, tall children from short adults — all to solve this almost-never-happens problem. This is no small ask, since without this requirement, the vehicles don’t even have to tell a dog from a crawling baby — they just know they should not run over anything roughly shaped like that.
We also have to understand that humans have so many accidents, that as a society we’ve come to just accept them as a fact of driving, and built a giant insurance system to arrange financial compensation for the huge volume of torts created. If we tried to resolve every car accident in the courts instead of by insurance, we would vastly increase the cost of accidents. In some places, governments have moved to no-fault claim laws because they realize that battling over something that happens so often is counterproductive, especially when from the standpoint of the insurers, it changes nothing to tweak which insurance company will pay on a case by case basis. In New Zealand, they went so far as to just eliminate liability in accidents, since in all cases the government health or auto insurance always paid every bill, funded by taxes. (This does not stop people having to fight the Accident Compensation Crown Corporation to get their claims approved, however.)
While the insurance industry total size will dwindle if robocars reduce accident rates, there are still lots of insurance programs out there that handle much smaller risks just fine, so I don’t believe insurance is going away as a solution to this problem, even if it gets smaller.
Submitted by brad on Sat, 2013-09-28 13:42.
I’ve written about the issues relating to robocars and walking before. On one hand, some people may find themselves hardly ever walking with convenient door-to-door robocar transportation. Others may find the robocars may enable walking by allowing one-way waking trips, or enabling trips that that allow drive-walk-drive (eliminating short driving trips done just to save the trouble of walking back to get the car.)
Some similar factors apply to cycling. In a lot of the world, people bike because it’s much cheaper and they can’t afford a car. In the richer countries, most people can afford cars, but people bike because they enjoy it, or seek the exercise. They may also wish to avoid traffic, take routes only bikes can take, or avoid burning gasoline.
Let’s consider something possible with robocars: on-demand bicycle delivery. This could either be small delivery robots which can hold bicycles, or the “bikebot” — a small robot that clamps onto a bicycle and uses the bike’s wheels in concert with the robot’s. The bikebot could be a very efficient way to delivery a bicycle — certainly using less energy per mile than a human being does, or that producing the human’s food does. (A future bike could be designed so that a bikebot module can be clamped to it easily.)
Bicycles on demand offer the chance to cycle just when you want to. This could increase by quite a bit the times when you actually would cycle.
This gets combined with the robocar’s one-way taxi ability for humans. The robocar can bring the humans, and/or the bicycles to the places they want to bike. (More efficiently, too, since bikes on bike racks are not very aerodynamic.)
Just like it does for walking, the multi-mode, multi-leg trip becomes enabled. For example, I often find I drive to Google, and then to Nasa which is 2.5 miles away, and then back home. The 2.5 mile leg is ideal for cycling — there’s even a bike trail for much of it — but I can’t do this. First I would have to always bring my bike. (While Google does provide bikes, they are tiny single-gear bikes not meant to leave campus.) I could do the round-trip to come back and get my car, but that’s less convenient and can hit some nasty traffic patterns — traffic in and out of Google at rush hour is very bad. This is a personal example, but I am sure you can all think of examples from your own life where you take an intermediary trip today (in your car) of a few miles that might be very nice for biking.
Weather is another impediment to biking. When I used to bike commute, I would drive instead if the forecast called for rain in the afternoon, even if it was nice in the morning. With robocars I could bike in, and ride back.
Bike delivery means choice of bike. Recumbents are hard to carry in cars, but no challenge for a custom bike delivery robot. They are more comfortable to ride and faster on flats and downhill. You could even climb a hill in a diamond bike and descend in a recumbent. Or, let’s face it, you could also have the cheater’s option of climbing the hard hills in a car, or with power assist, and riding alone when going down or traversing flatter terrain. That might be a cheater’s option, but it would get more people cycling. Your gear could shadow you in a small cargo robot.
The robocar also offers easy transportation for you, and your bikes, to the places where it’s fun to bike. Get driven to the coast, then bike it, then get driven back from your endpoint. Or bike the “interesting parts” and drive the boring (or difficult) ones.
The main issue? At least at first, a human will need to be there to put a bike into a bike-delivery robot or clamp a bikebot on the bike. That means you must declare your destination in advance, with enough time to get that robot to that spot so you can hand over the bike. Perhaps in the future, there will be robots that can pick up a waiting bike without a human to help. Quick one-way trips will probably not be with your personal bike, but rather a rental. While there are those who insist on their personally chosen bike for long rides, most people can tolerate a quality rental bike for a quick urban leg. Trikes, which are super easy to ride, can also be offered, and even bikes and trikes with motor assist when you want the non-exercise advantages of a bicycle could be provided.
Cycling could also be great for commute times. Many commuters might be happy to get a ride (perhaps even in groups) to the outskirts of the CBD, but as they enter the congested zone, have their car drop them off next to a bike for a quick ride to work. Long enough to get some exercise but not long enough to need a shower. This does present a problem when it rains and everybody wants to ride all the way in, though.
In the less developed world, where the bicycle is the transportation of choice due to cost, the robocar will take away some riders as it offers lower-cost transportation, protected from the weather, without up-front investment. However, eventually the above factors from the developed world will bring people back to the bicycle even though they can afford the car.
Submitted by brad on Thu, 2013-09-26 18:45.
I recently read a complaint by an EV driver that the charging station at De Anza College cost 55 cents/kwh. The national average price for electricity is around 10 cents, and at that price a typical electric car costs under 3 cents/mile for electricity. Gasoline costs about 8 cents/mile in a Prius, about 13 cents in a decent non-hybrid and 18 cents/mile in the average car which gets 22mpg. (At least here in California.) But the college’s charger’s electricity is almost 15 cents/mile in most electric sedans today, which is more than the gasoline in any gasoline car an eco-conscious person is likely to buy. (California Tier III electricity is 30 cents/kwh and thus almost as much.)
The price of charging stations varies wildly. A lot of them are free still, financed by other motivations. Tesla’s superchargers are free — effectively part of the cost of the car. It’s not uncommon for parking lots to offer free charging if you pay for parking, since parking tends to cost a fair bit more. After all, you won’t put more than 20kwh in a Leaf (and probably a lot less) and that costs just $2 at the average grid price.
This got me thinking of how the economics of charging will work in the future when electric cars and charging stations are modestly plentiful. While the national grid average is 10 cents, in many places heavy users can pay a lot more, though there are currently special deals to promote electric cars. Often the daytime cost for commercial customers is quite a bit higher, while the night is much lower. Charging stations at offices and shops will do mostly day charging; ones in homes and hotels will do night charging.
Unlike gasoline pumping, which takes 5 minutes, charging also involves parking. This is not just because charging takes several hours, but because that is enough time that customers won’t want to come and move their car once full, and so they will take the space for their full parking duration, which may be 8 or more hours.
Charging stations are all very different in utility. While every gasoline station near your route is pretty much equivalent to you, your charging station is your parking spot, and as such only the ones very close to your destination are suitable. While a cheap gas station 2 miles off your route would have a line around the block, a free charging stations 2 miles away from your destination is not that attractive! More to the point, the charging point close to your destination is able to command a serious premium. That have a sort of monopoly (until charging stations become super common) on charging at the only location of value to you.
Put another way, when buying gasoline, I can choose from all the stations in town. When picking an EV charge, I can only choose from stations with an available spot a short walk from my destination. Such a monopoly will lead to high prices in a market where the stations are charging (in dollars :-) what the market will bear.
The market will bear a lot. While the electricity may be available cheap, EV owners might be easily talked into paying as much for electricity as gasoline buyers do, on a per-mile basis. The EV owners will be forgetting the economics of the electric car — you pay the vast bulk of your costs up front for the battery, and the electrical costs are intended to be minor. If the electricity cost rivals that of gasoline, the battery cost is now completely extra.
Naturally, EV owners will do at least half their charging at home, where they negotiate the best rate. But this could be worse, as they might well be talked into looking at the average. They could pay 80 cents/kwh in the parking lot and 10 cents/kwh at home, and figure they are getting away with 45 cents and “still beating gasoline.” They would be fooling themselves, but the more people willing to fool themselves, the higher prices will go.
There is another lack of choice here. For many EV drivers, charging is not optional. Unless they have easy range to get back home or to another charging place they will spend lots of time, you must charge if you are low and the time opportunity presents itself. To not do so is either impossible (you won’t get home) or very foolish (you constrain what your EV can do.) When you face a situation where you must charge, and you must charge in a particular place, the potential for price gouging becomes serious. read more »
Submitted by brad on Wed, 2013-09-18 21:24.
It began with reports on a job ad at Tesla for an ADAS engineer to work on self-driving systems, and now there is a declaration from Elon Musk of a desire for a semi-automated car in three years. Musk says he expect the car to be “90% automated” which I will interpret as meaning it does highway driving. It is not said if this is the same sort of highway driving found in products like Cadillac’s “super cruise” or similar offerings from BMW, Mercedes, Volvo and others — which requires the driver be alert and watching, or a much harder full cruise ability that allows the driver to do other things, like read. I’m pretty sure it’s not a car that can run unmanned — Musk correctly feels that is a whole lot extra.
My reaction to this is mixed, in that there are things that make sense and don’t make sense.
On the plus side:
- Tesla is a great car company, and as a brand new one, perhaps the one most capable of not thinking like a car company. This is a big advantage. There is already a great culture of car innovation there.
- Tesla has a focus on great and novel car experiences, regardless of price, and this fits in well with that. Their customers will not be bothered by the initial high cost of the hardware.
- Their cars are already pretty much drive-by-wire and easy to adapt.
- If Tesla does decide to work with Google (the articles say they will not) there is already a strong friendship between the two CEOs
- Even in the best car, there are certainly lots of roads where you would rather not do the driving.
- With inductive charging (or some fancy plugging-in robot) it’s possible the car could do some self-parking and more importantly, self-recharging.
On the negative:
- Tesla’s cars are hugely fun to drive. While I believe for every car there value in having it drive itself on many roads, I would have to say the Teslas are the cars for which this is the least true! So it’s not that one would not appreciate self-driving in one’s Tesla, but it’s just that you would appreciate it even more in almost all other cars.
- Electric cars are not currently suitable as taxis that drive all-day, though Tesla has talked about battery swap, which would solve that issue. I doubt they mean to sell them for that market, as they would not be self-delivering in any event.
- Teslas are unjustifiably expensive. Well, unjustifiable to other than early adopters or those who just want the best at almost any price. That may change as batteries drop in price, though.
- If this is just super-cruise where you must pay attention, it’s nice, but not a revolution. Not yet, anyway.
Submitted by brad on Thu, 2013-09-12 16:50.
Videos have been released on some real-world tests of robocars. The most notable is from Mercedes.
As a nice reflection on the past, Mercedes drove the 100km route done by Bertha Benz in the first automotive road trip 125 years ago. You will also find that this alternate video is much better at talking about the technical details of the vehicle.
The Vislab team from Parma also released video of their drive around town. As the name suggests, Vislab’s research has a focus on computer vision, though this test vehicle also has 3 small LIDARs.
The Mercedes video has a lot of statements from MB engineer Ralf Herrtwich about their goals in doing this using existing sensors (cameras and radar primarily) and not (though he does not name it) LIDAR which he says is years or decades away. While I don’t want to criticise the accomplishments of his team, nor in any way deny that everybody would love to be able to make a safe driving system using the most cost-effective sensors, his philosophy seems backwards to me.
First, those of us used to Moore’s law think that planning to use hardware that is expensive today but which will come down greatly in price by the time things are commercialized is the obviously right strategy. It seems backwards to limit yourself to the technologies of today in planning a product for the future just because they are cheaper today. To use the metaphor of a great Canadian athlete, you skate to where the puck is going to be.
This is magnified by the fact that the problems of robocars are safety problems, not problems of cost or ones of appearance. With safety as the dominant goal, it seems very odd to me to imagine that one would, in the first vehicles to be made, avoid using a sensor that could improve safety and performance markedly just because of cost or appearance. If the cost difference were forecast to be ridiculous, one could consider it, but it makes no sense if the cost is within the noise to early adopters. That’s why Tesla is able to succeed with such an expensive car — the early adopters are more interested in a cool, high-performance electric car than they are in the cost. The other argument that is made — that the established sensors are more tested and robust — has some merit but is surely a short term optimization.
It could be argued that attempting to build a vehicle without LIDAR is skating to where the puck is going to be in the next game. After all, there is optimism that vision and radar will be enough for safe driving some day. As we all know, humans can drive with simple vision — even with one eye closed — and no radar or other sensors aside from hearing. So some day, cameras and a lot of processing probably can safely drive a car, and do it with low cost hardware. But the first production robocar? Deliberately not having lasers when it’s such a challenge to meet the safety goals? It seems very unlikely.
The notes on appearance are also odd to me. (It is commonly noted that research sensors like the Velodyne are big and make the car look unusual and not like a car.) We even see the IEEE Spectrum keen on how the new CMU car does not look like a robot) unlike BOSS from the urban challenge. While the research vehicles like BOSS were over-the-top on top, I think the reaction of early adopters is going to be quite the opposite. They will want their shiny new robocar to look distinctive and clearly different from regular cars. Prius owners reacted the same way, and there was not even much need for the Prius to have such a distinctive shape, though being more like a raindrop never hurts.
I suspect this approach is in part inspired by a marketing goal. The auto companies, not wanting to appear to be trailing Google on robocar research, are making extra effort to appear to be on a different course, and in fact ahead of Google and the rest on that path. “We’re doing what the competition is doing, but we’re not as far along” is not a very good press release. That’s OK if it were just for appearances — and I’m in favour of there being many competing approaches because any paradigm, including mine, can turn out to be wrong — but I hope that these teams really expect their approach is the best and fastest path to a safe and capable vehicle.
Here, by the way, are more details of the 33 mile trip by the GM/CMU collaboration. This vehicle has an “automotive grade” LIDAR — meaning one of the smaller ones that is one to four planes, not the giant 64 plane Velodyne used by CMU’s BOSS, Google and many others.
Submitted by brad on Sun, 2013-09-08 13:17.
Recent news has been big. First, Nissan announced it would sell robocars by 2020 and now today Daimler has announced the same. (Note that the 2014 S-class is the first car with a self-driving feature, a “you must still pay attention” traffic jam autopilot.)
In addition, sources have claimed that Google is either about to announce a collaboration with Conti on Sept 12 or is making plans to produce its own car and taxi service. (I was quoted, though not about Google, in one of the artciles in the series.
While I don’t comment on Google’s plans, I do believe it has one big advantage in this race. It doesn’t know what the rules of the car industry are, and has no desire to follow them. The car companies have huge resources, and better expertise on cars, but their internal rules and practices, honed over a century, are sure to hobble them. They won’t take the risks that non-car companies will take, won’t want to damage existing business lines, and will face attacks within the companies from the “company immune system” which seeks to attack disruptive ideas within big companies.
Google’s main impediment is that it is also a big company, though an unusual one. But this business is so hard to enter that we have yet to see a start-up make a play.
The statements from all these parties will do lots of good, lighting fires under the other players, including the unannounced ones. I believe that in the 2020s, the software and sensor system which drives the car will be the most important part of the car, more important than even the engine. While the world will be better off if there are multiple competing suppliers of this part, whoever dominates this will dominate the car industry.
Submitted by brad on Mon, 2013-08-19 16:59.
Probably the most expensive add-on that people get in their cars today is the stereo. Long ago, cars often came without stereos and there was a major aftermarket. The aftermarket is still here but most people elect for factory stereos which fit in seamlessly with the car and often cost a huge amount of money.
The car’s not a great place to listen to music — it’s noisy and you are distracted and you often stop and have to get out in the middle of a song. But because people find they listen to more music in their cars than at home, they often pay huge bucks for a fancy car stereo. (Not counting the people who deliberately buy a system so loud it’s meant for other people outside the car to hear.)
While you could put a nice stereo system in a robocar, and some people will, another way they can save money is they don’t need to have much audio at all, not once they can do full-auto operation. The prohibition on headphones by the driver should go away, and it could become popular to just use nice headphones — possibly noise cancelling headphones or in-ear noise-blocking phones. A better audio experience with much less noise, and a lot cheaper too. And there is the option for each person in the car to have their own headphones and tune their own audio stream.
People will like to share, so the car might contain a simple audio distribution system to feed audio streams to people who are sharing, though the source of the music should still be somebody’s phone or device, not something built into the car. In addition, there could be a system to mix in some of the in-cabin audio, so you can still hear the other people when they talk. Microphones on each person’s headphones could pick up their voices and actually provide a clearer read of their voices. Headphones with position sensors could allow simulation of stereo on the other people. Alternately a microphone array could exist around the car, particularly at each seat.
There are some downsides to push things into the traditional way:
- Wearing headphones is uncomfortable on long trips
- They are a pain to remember to put on. You want to avoid cords, so they would be wireless, but then you must be sure to put them in their charging dock.
- On small aircraft, there is so much noise that everybody does it this way, but they tend to be bulky (due to the high noise) and unpopular for that reason
So people might elect to still have decent speakers and listen to music without headphones. But there is less need to buy a really expensive sound system, since if you want the top quality you probably want to go for the headphones. This may also apply to decisions to do expensive sound elimination in the car. For some, nothing may change, but that’s OK. What’s interesting is the option to do car sound in ways never done before.
Submitted by brad on Wed, 2013-08-14 13:03.
Frequently, in reporting on robocars, it is often cited that one of their key benefits will be the way they enable car sharing, greatly reducing the number of cars that need to exist to serve the population. It is sometimes predicted that we’ll need to make fewer cars, which is good for the environment.
It is indeed true — robotaxi service, with cars that deliver themselves and drop you off, does greatly enable car sharing. But from the standpoint of modern car sharing, it may enable it too well, and we may end up having to manufacture more cars, not fewer.
Today’s car sharing companies report statistics that they replace around 13 privately owned cars for every car in the carsharing fleet. Some suggest it’s even as high as 20.
This number is impossible for average drivers, however. The average car is driven 12,000 miles/year. To replace 13 average cars would require a vehicle that was actively driving, not just signed out, 11 hours/day and each vehicle would wear out in 1-2 years.
Three things are happening.
- Carsharing is replacing the more marginal, less used vehicles. A household replaces a 2nd or 3rd car. Carsharing is almost always used by people who do not commute by car.
- Carsharing is often considerably less convenient than a private car. It discourages driving, pushing its users into other modes of transport, or selecting for customers who can do that.
- Related to that, carsharing shows the true cost of car ownership and makes it incremental. That cost is around $20/hour, and people rethink trips when they see the full cost laid out per mile or per hour. With private cars, they ignore most of the cost and focus only on the gasoline, if that.
The “problem” with robocars is that they’re not going to be worse than having a private car. In many ways they will be better. So they will do very little of the discouragement of car use caused by present day carshare models. The “dark secret” of carsharing is that it succeeds so well at replacing cars because of its flaws, not just its virtues.
Robotic taxis can be priced incrementally, with per-mile or per-hour costs, and these costs will initially be similar to the mostly unperceived per-mile or per-hour costs of private car ownership, though they will get cheaper in the future. This revelation of the price will discourage some driving, though robotaxi companies, hoping to encourage more business, will likely create pricing models which match the way people pay for cars (such as monthly lease fees with only gasoline costs during use) to get people to use more of the product.
There is an even stronger factor when it comes to robotaxis. A hard-working robotaxi will indeed serve many people, and as such it will put on a lot of miles every year. It will thus wear out much faster, and be taken out of service within 4-5 years. This is the case with today’s human driven taxicabs, which travel about 60,000 miles/year in places like New York.
The lifetime of a robotaxi will be measured almost exclusively in miles or engine-hours, not years. The more miles people travel, the more vehicles will need to be built. It doesn’t matter how much people are sharing them.
The core formula is simple.
Cars made = Vehicle Miles Travelled (VMT) / Car lifetime in miles
The amount of sharing of vehicles is not a factor in this equation, other than when it affects VMT.
Today the average car lasts 200,000 miles in California. To be clear, if you have 8,000 customers and they will travel two billion miles in 20 years (that’s the average) then they are going to need 8,000 cars over those years. It almost doesn’t matter if you serve them with their own private car, and it lasts all 20 years, or if you get 2,000 cars and they serve 4 people each on average and wear out after 5 years. read more »
Submitted by brad on Mon, 2013-08-05 12:38.
Our technology is having trouble with settling on a name. That’s OK before it’s mainstream but will eventually present a problem. When people in the field are polled on what name they like, there is no clear winner. Let’s look at some of the commonly used candidates:
Recently, this has become the most common term used in the press. There is a “Driverless Car Summit” and the Wikipedia page has used that name for some time.
In spite of this popularity, the term is very rarely used by people actually building the vehicles. Attendees at the “Driverless Car Summit” when polled all said they dislike it. Until recently, the most common news story about a driverless car would say, “then the driverless car rolled down the hill and careened into the other lane, hitting a tree.”
My personal view is that this term is like “horseless carriage.” Long ago the most remarkable thing about the automobile was that it had no horse. Here it’s the lack of driver (or at least lack of action by the driver.) Of course, these cars have something driving them, but it’s a computer system. While this term is most popular, I am confident it will fade away and seem quaint, like horseless carriage did.
This term is popular among developers of the cars. Its main problem is that it’s too long to be a popular term. The acronym SDC is a reasonable one. In web hits, this is tied with Driverless Cars, but falls behind that name in searches and news mentions.
This term was most popular in the early years, though it is most commonly found in research environments and in the military sphere. In the military they also use “unmanned ground vehicle” — another term too unwieldy for the public —though they usually refer to remote controlled vehicles, not self-driving ones.
Annoyingly, the acronym “AV” has another popular meaning today. Most of the terms here are too long to become common use terms, and so will be turned into acronyms or shortened, but this one has an acronym problem.
Automated Road Vehicle
This term has minor traction, almost entirely due to the efforts of Steve Shladover of UC Berkeley. In his view, the word autonomous is entirely misused here and the correct term is automated. Roboticists tend to differ — they have been using “autonomous” to mean “not remote controlled” for many years. There are two meanings of autonomous in common use. One is to be independent of direct control (which these cars are) and the other one, “self-governing” is the one Steve has the issue with. As a member of the program committee for TRB’s conference on the area, he has pushed the “automated” name and given it some traction.
Unfortunately, to roboticists, “automated” is how you describe a dishwasher or a pick-and-place robot; it’s a lower level of capability. I don’t expect this terminology to gain traction among them.
I selected this term for these pages for a variety of reasons. It was already in modest use thanks to a Science Channel documentary on the DARPA challenge called “robocars.”
- Talking to teams, they usually just called their vehicle “the robot” or “the car.”
- It is short, easy to say, and clear about what it means
- It is distinct and thus can easily be found in online searches
- It had some amount of existing use, notably as the title of a documentary on the Science Channel about the DARPA challenges
However, it is doing poorly in popularity and only has about 21,000 web pages using it, so I may need to switch away from it as well if a better term appears. Today it reminds people too much of robotics, and the trend is to move away from that association.
On the other hand, no other term satisfies the criteria above, which I think are very good criteria. read more »
Submitted by brad on Fri, 2013-08-02 10:52.
I’m often asked whether robocars will keep themselves to the speed limit and refuse to go faster, unlike cruise controls which let the driver set the automated speed. In many countries, the majority of human drivers routinely exceed the limit which could present issues. On the other hand, vendors may fear liability over programming their cars to do this, or even programming them to allow their human overlord to demand it.
While the right answer is a speed-limit doctrine like the French Autoroute, where the limit is 130 kph/80 mph and few disobey it, until we can come to that answer, the math suggests that travel might be overall safer if the robocars are allowed to speed in the same way humans do, at the request of humans. And indeed, that is how prototype implementations have been built.
I felt this subject (and related subjects about how cars should deal with laws that are routinely broken by human drivers) deserved a special article. Read about it at:
Robocars and the Speed Limit
Submitted by brad on Mon, 2013-07-22 16:08.
A nice result for Vislab of Parma, Italy. They have completed a trial run on public roads using their mostly vision-based driving system. You can see a report on the Vislab site for full details. The run included urban, rural and highway streets. While the press release tries to make a big point that they did this with a vacant driver’s seat, the video shows a safety driver in that seat at all times, so it’s not clear how the test was done. They indicate that the passenger had an emergency brake, and a chase car had a remote shutoff as well.
The Vislab car uses a LIDAR for forward obstacle detection, but their main thrust is the use of cameras. An FPGA-based stereo system is able to build point clouds from the two cameras. Driving appears to have been done in noonday sunlight. (This is easy in terms of seeing things but hard in terms of the harsh shadows.)
The article puts a focus on how the cameras are cheaper and less obtrusive. I continue to believe that is not particularly interesting — lasers will get cheaper and smaller, and what people want here is the best technology in the early adopter stages, not the cheapest. In addition, they will want it to look unusual. Cheaper and hidden are good goals once the cars have been deployed for 5-10 years.
This does not diminish the milestone of their success, making the drive with this sensor set and in these conditions.
Submitted by brad on Thu, 2013-07-18 19:56.
This week I attended the Transportation Research Board Workshop on Automated Road Vehicles which has an academic focus but still has lots of industry-related topics. TRB’s main goal is to figure out what various academics should be researching or getting grants for, but this has become the “other” conference on robocars. Here are my notes from it.
Bryant Walker Smith told of an interesting court case in Ontario, where a truck driver sued over the speed limiter put in his truck and the court ruled that the enforced speed limiter was a violation of fundamental rights of choice. One wonders if a similar ruling would occur in the USA. I have an article pending on what the speed limit should be for robocars with some interesting math.
Cliff Nass expressed skepticism over the ability to have easy handover from self-driving to human driving. This transfer is a “valence transfer” and if the person is watching a movie in a tense scene that makes her sad or angry, she will begin driving with that emotional state. More than one legal scholar felt that quickly passing control to a human in an urgent situation would not absolve the system of any liability under the law, and it could be a dangerous thing. Nass is still optimistic — he notes that in spite of often expressed fears, no whole field has been destroyed because it caused a single fatality.
There were reports on efforts in Europe and Japan. In both cases, government involvement is quite high, with large budgets. On the other hand, this seems to have led in most cases to more impractical research that suggests vehicles are 1-2 decades away.
Volkswagen described a couple of interesting projects. One was the eT! — a small van that would follow a postman around as he did his rounds. The van had the mail, and the postman did not drive it but rather had it follow him so he could go and get new stacks of mail to deliver. I want one of those in the airport to have my luggage follow me around.
VW has plans for a “traffic jam pilot” which is more than the traffic jam assist products we’ve seen. This product would truly self-drive at low speeds in highway traffic jams, allowing the user to not pay attention to the road, and thus get work done. In this case, the car would give 10 seconds warning that the driver must take control again. VW eventually wants to have a full vehicle which gives you a 10 minute warning but that’s some distance away. read more »
Submitted by brad on Thu, 2013-07-11 21:06.
The Vislab team from Parma, Italy, which you may remember did the intermittently autonomous drive from Italy to Shanghai a couple of years ago is back with a new vehicle, dubbed BRAiVE which tomorrow begins testing on real urban streets.
The difference is this car is mostly based on vision systems, the specialty of Vislab. You can see a photo gallery of the car but it deliberately does not look particularly different. You can see a few low profile sensors. They claim the car uses “mostly cameras” so it’s not clear if there is still a LIDAR on the vehicle or it’s just cameras and radar. The cars to Shanghai used an array of both cameras and single plane LIDARs. It is said that the sensors are “low cost” though an exact list is not given.
This will be an interesting experiment. Previous vision based systems have not proven adequate for urban driving. They have been able to do it but not reliably enough to trust people’s lives to it. Cameras remain attractive for their low cost and other reasons outlined in my recent article on LIDAR vs. cameras.
The sensors on this vehicle are not that obvious. There remain two schools of thought on this. One believes that a significant change in the car form factor with obvious sensors will be a turn-off for buyers. Others think buyers, especially early adopters, will actually consider unusual looking sensors a huge plus, wanting the car to stand out. I’m in the latter camp, and think the Prius is evidence of this. Its unusual shape outsells all other hybrids combined, even the more ordinary looking Camry hybrid, where the Camry is the best selling car there is. However, there will be markets for both designs.
It will be interesting to see the results of this research, and what rates of accuracy they gain for their vision system. Lots of competing approaches is good for everybody.
Submitted by brad on Wed, 2013-06-19 11:14.
The AUVSI summit on “driverless” cars last week contained 2 days of nothing but robocars, and I reported on issues regarding Google and policy in part 1.
As noted, NHTSA released their proposal for how they want to regulate such vehicles. In it, they defined levels 0 through 4. Level 2 is what I (and GM) have been calling “super cruise” — a car which can do limited self driving but requires constant human supervision. Level 3 is a car which can drive without constant attention, but might need to call upon a human driver (non-urgently) to handle certain streets and situations. Level 4 is the fully automatic robocar.
Level 2 issues
Level 2 is coming this year in traffic jams in the Mercedes S and the BMW 5, and soon after from Audi and Volvo. GM had announced super cruise for the 2015 Cadillac line but has pulled back and delayed that to later in the decade. Nonetheless the presentation from GM’s Jeremy Salinger brought home many of the issues with this level.
GM has done a number of user studies in their super cruise cars on the test track. And they learned that the test subjects very quickly did all sorts of dangerous things, definitely not paying attention to the road. They were not told what they couldn’t do, but subjects immediately began texting, fiddling around in the back and even reading (!) while the experimenters looked on with a bit of fear. No big surprise, as people even text today without automatic steering, but the experimental results were still striking.
Because of that GM is planning what they call “countermeasures” to make sure this doesn’t happen. They did not want to say what countermeasures they liked, but in the past, we have seen proposals such as:
- You must touch the wheel every few seconds or it disengages
- A camera looks at your eyes and head and alerts or disengages if you look away from the road for too long
- A task for your hands like touching a button every so often
The problem is these countermeasures can also get annoying, reducing the value of the system. It may be the lack of ability to design a good countermeasure is what has delayed GM’s release of the product. There is a policy argument coming up about whether level 2 might be more dangerous than the harder levels 3 and above, because there is more to go wrong with the human driver and the switches between human and machine driving. (Level 4 has no such switches, level 3 has switches with lots of warning.)
On the plus side, studies on existing accidents show that accident-avoidance systems, even just forward collision avoidance, have an easy potential for huge benefits. Already we’re seeing a 15% reduction in accidents in some studies just from FCA, but studies show that in 33% of accidents, the brakes were never applied at all, and only in just 1% of accidents were the brakes applied with full force! As such, systems which press the brakes and press them hard when they detect the imminent accident may not avoid the accident entirely, but they will highly reduce the severity of a lot of accidents. read more »
Submitted by brad on Sun, 2013-06-16 09:38.
I was sadly informed this morning by Ann Lowson that transportation pioneer Martin Lowson has fallen to a stroke this weekend.
Martin had an amazing career but it was more amazing that he was still actively engaged at age 75. We shared a panel last month in Phoenix at the people-mover conference and continued our vigourous debate on the merits of cars like his on closed guideways compared to robocars.
His career included leading a large team on the Apollo project, and building the world’s fastest helicopter, as well as faculty positions at Bristol, and you can read some about it here. For me, his big contribution was to found the ULTra PRT company, the first to commercially deploy a PRT. It runs today at Heathrow, moving people between the terminal and the business parking lot.
PRT was conceived 50 years ago, and many, including Martin and myself, were fascinated by the idea. More recently, as readers know, I decided the PRT vision of personal transportation could be realized on city streets by robocars. It’s easier to do it today on dedicated guideway, but the infrastructure costs tell me the future lies off the guideway.
That doesn’t diminish the accomplishment of being the first to make it work on the guideway. ULTra uses small cars on rubber tires, not a train on rails. They are guided by a laser rangefinder and are fully automated, with no steering wheel.
Last year I invited Martin in to give a talk to Google’s car team, and he got a ride in the car, which he quite enjoyed, even though it didn’t convince him that they were the future. But unlike other skeptics, I gave him the deepest respect for his skill and experience. People who can found companies and lead engineering and public acceptance breakthroughs while senior citizens are a very rare thing, and the world will miss him.
Submitted by brad on Sat, 2013-06-15 15:43.
This week I attended AUVSI’s “Driverless Car Summit” in Detroit. This year’s event, the third, featured a bigger crowd and a decent program, and will generate more than one post.
I would hardly call it a theme, but two speakers expressed fairly negative comments about Google’s efforts, raising some interesting subjects. (As an important disclaimer, the Google car team is a consulting client of mine, but I am not their spokesman and the views here do not represent Google’s views.)
The keynote address came from Bryan Reimer of MIT, and generated the most press coverage and debate, though the recent NHTSA guidelines also created a stir.
Reimer’s main concern: Google is testing on public streets instead of a test track. As such it is taking the risk of a fatal accident, from which the blowback could be so large it stifles the field for many years. Car companies historically have done extensive test track work before going out on real streets. I viewed Reimer’s call as one for near perfection before there is public deployment.
There is a U-shaped curve of risk here. Indeed, a vendor who takes too many risks may cause an accident that generates enough backlash to slow down the field, and thus delay not just their own efforts, but an important life-saving technology. On the other hand, a quest for perfection attempts what seems today to be impossible, and as such also delays deployment for many years, while carnage continues on the roads.
As such there is a “Goldilocks” point in the middle, with the right amount of risk to maximize the widescale deployment of robocars that drive more safely than people. And there can be legitimate argument about where that is.
Reimer also expressed concern that as automation increases, human skill decreases, and so you actually start needing more explicit training, not less. He is as such concerned with the efforts to make what NHTSA calls “level 2” systems (hands off, but eyes on the road) as well as “level 3” systems (eyes off the road but you may be called upon to drive in certain situations.) He fears that it could be dangerous to hand driving off to people who now don’t do it very often, and that stories from aviation bear this out. This is a valid point, and in a later post I will discuss the risks of the level-2 “super cruise” systems.
Maarten Sierhuis, who is running Nissan’s new research lab (where I will be giving a talk on the future of robocars this Thursday, by the way) issued immediate disagreement on the question of test tracks. His background at NASA has taught him that you “fly where you train and train where you fly” — there is no substitute for real world testing if you want to build a safe product. One must suspect Google agrees — it’s not as if they couldn’t afford a test track. The various automakers are also all doing public road testing, though not as much as Google. Jan Becker of Bosch reported their vehicle had only done “thousands” of public miles. (Google reported a 500,000 mile count earlier this year.)
Heinz Mattern, research and development manager for Valeo (which is a leading maker of self-parking systems) went even further, starting off his talk by declaring that “Google is the enemy.” When asked about this, he did not want to go much further but asked, “why aren’t they here? (at the conference)” There was one Google team employee at the conference, but not speaking, and I’m not am employee or rep. It was pointed out that Chris Urmson, chief engineer of the Google team, had spoken at the prior conferences. read more »
Submitted by brad on Sun, 2013-06-09 18:13.
I’m off for AUVSI’s “Driverless Car Summit” in Detroit. I attended and wrote about last year’s summit, which, in spite of being put on by a group that comes out of the military unmanned vehicle space, was very much about the civilian technology. (As I’ve said before, I have a dislike for the term “driverless car” and in fact at the summit last year, the audience expressed the same dislike but could not figure out what the best replacement term was.)
I’ll be reporting back on events at the summit, and making a quick visit to my family in Toronto as well. I will also attend the Transportation Research Board’s conference on automated vehicles at Stanford in July.
Then I’m back for the opening of our Singularity University Graduate Studies Program for 2013 at NASA Ames Research Park this coming weekend. My students will get some fun lectures on robocars, as well as many other technologies. Early bird tickets for the opening ceremony are still available.
On June 20, I will give a talk at a meeting of the new Silicon Valley Autonomous Vehicle Enthusiasts group. This group has had one talk. The talks are being hosted at Nissan’s new research lab in Silicon Valley, where they are researching robocars. I just gave a 10 minute version of my talk at Fujitsu Labs’ annual summit last week, this will be the much longer version!
SU consumes much of my summer. In the fall, you’ll see me giving talks on Robocars and other issues in Denmark, London, Milan as well as at our new Singularity University Summit in Budapest in November, as well as others around the USA.
Submitted by brad on Thu, 2013-06-06 14:20.
There have been a wide variety of announcements of late giving the impression that somebody has “solved the problem” of making a robocar affordable, usually with camera systems. It’s widely reported how the Velodyne LIDAR used by all the advanced robocar projects (including Google, Toyota and many academic labs) costs $75,000 (or about $30,000 in a smaller model) and since that’s more than the cost of the car, it is implied that is a dead-end approach.
Recent stories include a ride in MobilEye’s prototype car by the New York Times, a number of reports of a claim from the Oxford team (which uses LIDAR today) that they plan to do it for just $150 and many stories about a Romanian teen who won the Intel science fair with a project to build a cheaper self-driving car.
I have written an analysis of the issues comparing LIDARS (which are currently too expensive, but reliable in object detection) and vision systems (which are currently much less expensive, but nowhere near reliable enough in object detection) and why different teams are using the different technologies. Central is the question of which technology will be best at the future date when robocars are ready to be commercialized.
In particular, many take the high cost of the Velodyne, which is hand-made in small quantities, and incorrectly presume this tells us something about the cost of LIDARs a few years down the road, with the benefits of Moore’s Law and high-volume manufacturing. Saying the $75,000 LIDAR is a dead-end is like being in 1982, and noting that small disk drives cost $3,000 and declaring that planning for disk drive storage of large files is a waste of time.
Cameras or Lasers in the robocar
I will add some notes about Ionut Budisteanu, the 19-year old Romanian. His project was great, but it’s been somewhat exaggerated by the press. In particular, he mistakenly calls LIDAR “3-D radar” (an understandable mistake for a non-native English speaker) and his project was to build a lower-cost, low-resolution LIDAR, combining it with cameras. However, in his project, he only tested it in simulation. I am a big fan of simulation for development, learning, prototyping and testing, but alas, doing something in simulation, particularly with vision, is just the first small step along the way. This isn’t a condemnation of Mr. Budisteanu’s project, and I expect he has a bright future, but the press coverage of the event was way off the mark.
Submitted by brad on Thu, 2013-05-30 13:42.
Today the National Highway Transportation Safety Agency (NHTSA) released their plan on regulation of automated vehicles, a 14 page document on various elements of the technology and how it might be regulated.
No regulations yet of course, but a message that they plan to look hard at the user interface, particularly on the handoff between a human driver and the system. All reasonable stuff. They define 4 levels of autonomy (similar to prior lists) and say they don’t expect full unmanned operation for some time, and discourage states from even making it legal to use level 3 (where the driver can do another task) by ordinary folks yet — only testing should be allowed.
It’s good that NHTSA is studying this, and they seem to understand that it’s too early to write regulations. It’s pretty easy to make mistakes if you write regulations before the technologists have even figured out what they intend. For example this document, as well as some Nevada law documents, suggested regulations that required the vehicle to hand over control if the driver used the wheel, brakes or accelerator. Yet by another logic, if the driver kicks the gas pedal by mistake and does not have her hands on the wheel, would we want the law to demand that the system relinquish the wheel, causing the vehicle to leave the lane if the driver doesn’t get on it quickly?
At this point their goal is lots of research on what to do, and that’s reasonable. Of course, the sooner the vehicles can get on the road safely, the sooner they can save lives, and NHTSA understands that. I also hope that the laws will not push small players out of the market entirely, as long as they can test and demonstrate safety as well as the bigger players.
Submitted by brad on Fri, 2013-05-24 17:06.
It seems that Better place has gone to… a better place to put it ironically. I’m not greatly surprised, I expressed my skepticism last year.
But I do believe in the idea of the self-driving electric taxi as the best answer for our future urban transportation. So how do you make it happen?
There’s a big problem with this vision. Electric cars today have perhaps 100 to 150 miles of range, which means 3 to 6 hours of operation, depending on the speeds you go. You can make more range (like a Telsa S) but only by adding a lot of weight and cost. But an effective taxi is on shift all day, or at least all the waking hours, and could easily operate 8 to 10 hours per day. While any taxi will have downtime during the day (particularly at off-peak hours) the recharge of the battery takes so long it’s hard to do during the day. Ideally you want to charge at night, when power is cheap. So let’s consider the options.
Large battery pack
You could make a vehicle with enough battery for a full day’s work, and charge it at night. This is very expensive today, and also takes a lot of room and weight in the vehicle, reducing its efficiency. You also need taxis at night so either way you have to have some taxis that work at night and recharge in the day, but not as many.
While battery swap did not pan out for Better Place, it actually makes much more sense for a robotic taxi fleet. You just need a few swap stations in the city. It doesn’t bother the robots if they take a while for a swap (mainly it bothers you because you need more stations.) And while humans would get very angry if they came to the swap station and saw they were 4th in line at 5 minutes/swap, robots would just schedule their swaps and get in and get out.
That’s all good, and it solves a few other problems. Taxis will be putting on lots of miles every day, and they will probably wear out their battery quickly. If the rest of the vehicle has not worn out, swap becomes ideal — replace the vehicle’s components when you need to, including the most expensive part, the battery. It also makes it easier to charge all batteries only at night, on cheap baseload power.
Swap also allows the batteries to be only used in the “easy” part of their duty cycle (from 80% to 20%) without much hassle. You only get 60% of the range, but you don’t care a lot, other than in having to
buy more battery packs. You just do the math on what’s cheaper.
A working supercharger that can recharge a vehicle in an hour solves the problem as well. Robotic taxis can always find a spare hour without much loss of efficiency. (Indeed with none as they will age by active mile or hour.) The big problem is that supercharging generally is felt to stress the batteries and reduce the lifetime of the packs. Certainly running a car on full cycle every day and supercharging it is not going to produce a happy battery.
A robocar supercharging station could do a few extra things, though. For example, you could hook the car up to a special cooling system to pump externally chilled coolant through the batteries, as heat is the big killer in the supercharge. You might even find a way to put some pressure in to keep the cases from expanding that much, as this is a big stressor when charging.
Supercharging probably has to be done in the day, with more expensive power. One charge for the morning peak and another for the evening. Some speculate it’s worth using your inventory of old battery packs to store power during the night to release in the day. Solar can also help during the day — on sunny days, at least.
While automated connection is good, you really would not have many supercharging centers, due to their high power needs, so they could have human staff to do the work.
Both the supercharging and battery swap stations do not need to be located all that conveniently for humans. Instead, you can put them near power substations where megawatts can be purchased.
More vehicles and ordinary L2 charging
If the batteries are more expensive than the vehicles, then perhaps just having more vehicles to house all your battery packs is the answer. Then you have vehicles spending their time idle, and charging at ordinary level 2 (6kw) rates. Full level 2 can add about 20 miles of range to a car like a Leaf in an hour. Depending on the usage patterns that might not be too bad. Of course it’s daytime power again, which is expensive. Urban taxis won’t go more than about 25mph on average if they are lucky, often less, particularly at rush hour. Suburban will go a bit faster. You need stations that allow a robot to recharge, which could mean inductive, or human-staffed, or eventual robotic plug-in systems. Don’t laugh at the idea of human-staffed. The robot will not be in a super rush, so stations near retail shops or existing gas stations would work fine as long as somebody can come out and tend to the robot on connect and disconnect within 5 minutes.
It may seem like more vehicles is more expensive, but that’s not necessarily true. It depends on how and why the vehicles wear out. Ideally you design the vehicle so battery and most major vehicle components all reach end-of-life at a similar time or that they can be replaced easily. That may mean a battery that can be swapped — but in the shop, not at an automatic swap station.
Plug in hybrids?
Plug in hybrids of course solve the problem, and they can charge when they can to be mostly electric and only use that gas engine more rarely. This actually creates a downside — it’s expensive to have a fossil fuel power train around to barely use it. And it adds a fair bit to the maintenance cost. This does allow for highway travel. Otherwise, you send a liquid fuel car to anybody wanting to do a long highway trip - save the electrics for the urban travel.
Very light vehicles
Today’s electrics use about 250 to 300 watt-hours/mile. OK, but not great. Efficient designs can go below 100 watt/hours per mile. That means doing 300 miles, which is enough for a full day in a city cab, needs only 30kwh (probably a 45kwh battery.) That’s a $22K battery today, but it will be a $9K battery by the end of this decade according to predictions. This might be quite reasonable.