A few weeks ago, in my article on myths I wrote why the development of “vehicle to vehicle” (V2V) communications was mostly orthogonal to that of robocars. That’s very far from the view of many authors, and most of those in the ITS community. I remain puzzled by the V2V plan and how it might actually come to fruition. Because there is some actual value in V2V, and we would like to see that value realized in the future, I am afraid that the current strategy will not work out and thus misdirect a lot of resources.
This is particularly apropos because recently, the FCC issued an NPRM saying it wants to open up the DSRC band at 5.9ghz that was meant for V2V for unlicenced wifi-style use. This has been anticipated for some time, but the ITS community is concerned about losing the band it received in the late 90s but has yet to use in anything but experiments. The demand for new unlicenced spectrum is quite appropriately very large — the opening up of 2.4gz decades ago generated the greatest period of innovation in the history of radio — and the V2V community has a daunting task resisting it.
In this series I will examine where V2V approaches went wrong and what they might do to still attain their goals.
I want to begin by examining what it takes to make a successful cooperative technology. History has many stories of cooperative technologies (either peer-to-peer or using central relays) that grew, some of which managed to do so in spite of appearing to need a critical mass of users before they were useful.
Consider the rise and fall of fax (or for that matter, the telephone itself.) For a lot of us, we did not get a fax machine until it was clear that lots of people had fax machines, and we were routinely having people ask us to send or receive faxes. But somebody had to buy the first fax machine, in fact others had to buy the first million fax machines before this could start happening.
This was not a problem because while one fax machine is useless, two are quite useful to a company with a branch office. Fax started with pairs of small networks of machines, and one day two companies noticed they both had fax and started communicating inter-company instead of intra-company.
So we see rule one: The technology has to have strong value to the first purchaser. Use by a small number of people (though not necessarily just one) needs to be able to financially justify itself. This can be a high-cost, high-value “early adopter” value but it must be real.
This was true for fax, e-mail, phone and many other systems, but a second principle has applied in many of the historical cases. Most, but not all systems were able to build themselves on top of an underlying layer that already existed for other reasons. Fax came on top of the telephone. E-mail on top of the phone and later the internet. Skype was on top of the internet and PCs. The underlying system allowed it to be possible for two people to adopt a technology which was useful to just those two, and the two people could be anywhere. Any two offices could get a fax or an e-mail system and communicate, only the ordinary phone was needed.
The ordinary phone had it much harder. To join the phone network in the early days you had to go out and string physical wires. But anybody could still do it, and once they did it, they got the full value they were paying for. They didn’t pay for phone wires in the hope that others would some day also pay for wires and they could talk to them — they found enough value calling the people already on that network.
Social networks are also interesting. There is a strong critical mass factor there. But with social networks, they are useful to a small group of friends who join. It is not necessary that other people’s social groups join, not at first. And they have the advantage of viral spreading — the existing infrastructure of e-mail allows one person to invite all their friends to join in.
Enter Car V2V
Car V2V doesn’t satisfy these rules. There is no value for the first person to install a V2V radio, and very tiny value for the first thousands of people. An experiment is going on in Ann Arbor with 3,000 vehicles, all belonging to people who work in the same area, and another experiment in Europe will equip several hundred vehicles. read more »
You’ve probably seen the battle going on between Elon Musk of Tesla and the New York Times over the strongly negative review the NYT made of a long road trip in a Model S. The reviewer ran out of charge and had a very rough trip with lots of range anxiety. The data logs published by Tesla show he made a number of mistakes, didn’t follow some instructions on speed and heat and could have pulled off the road trip if he had done it right.
Both sides are right, though. Tesla has made it possible to do the road trip in the Model S, but they haven’t made it easy. It’s possible to screw it up, and instructions to go slow and keep the heater low are not ones people want to take. 40 minute supercharges are still pretty long, they are not good for the battery and it’s hard to believe that they scale since they take so long. While Better Place’s battery swap provides a tolerable 5 minute swap, it also presents scaling issues —
you don’t want to show up at a station that does 5 minute swaps and be 6th in line.
The Tesla Model S is an amazing car, hugely fun to drive and zippy, cool on the inside and high tech. Driving around a large metro area can be done without range anxiety, which is great. I would love to have one — I just love $85K more. But a long road trip, particularly on a cold day? There are better choices. (And in the Robocar world when you can get cars delivered, you will get the right car for your trip delivered.)
Electric cars have a number of worthwhile advantages, and as battery technologies improve they will come into their own. But let’s consider the economics of a long range electric. The Tesla Model S comes in 3 levels, and there is a $20,000 difference between the 40khw 160 mile version and the 85kwh 300 mile version. It’s a $35K difference if you want the performance package.
The unspoken secret of electric cars is that while you can get the electricity for the model S for just 3 cents/mile at national grid average prices (compared to 12 cents/mile for gasoline in a 30mpg car and 7 cents/mile in a 50mpg hybrid) this is not the full story. You also pay, as you can see, a lot for the battery. There are conflicting reports on how long a battery pack will last you (and that in turn varies on how you use and abuse it.) If we take the battery lifetime at 150,000 miles — which is more than most give it — you can see that the extra 45kwh add-on in the Tesla for $20K is costing about 13 cents/mile. The whole battery pack in the 85kwh Telsa, at $42K estimated, is costing a whopping 28 cents/mile for depreciation.
Here’s a yikes. At a 5% interest rate, you’re paying $2,100 a year in interest on the $42,000 Tesla S 85kwh battery pack. If you go the national average 12,000 miles/year that’s 17.5 cents/mile just for interest on the battery. Not counting vehicle or battery life. Add interest, depreciation and electricity and it’s just under 40 cents/mile — similar to a 10mpg Hummer H2. (I bet most Tesla Model S owners do more than that average 12K miles/year, which improves this.)
In other words, the cost of the battery dwarfs the cost of the electricity, and sadly it also dwarfs the cost of gasoline in most cars. With an electric car, you are effectively paying most of your fuel costs up front. You may also be adding home charging station costs. This helps us learn how much cheaper we must make the battery.
It’s a bit easier in the Nissan LEAF, whose 24kwh battery pack is estimated to cost about $15,000. Here if it lasts 150K miles we have 10 cents/mile plus the electricity, for a total cost of 13 cents/mile which competes with gasoline cars, though adding interest it’s 19 cents/mile — which does not compete. As a plus, the electric car is simpler and should need less maintenance. (Of course with as much as $10,000 in tax credits, that battery pack can be a reasonable purchase, at taxpayer expense.) A typical gasoline car spends about 5 cents/mile on non-tire maintenance.
This math changes a lot with the actual battery life, and many people are estimating that battery lives will be worse than 150K miles and others are estimating more. The larger your battery pack and the less often you fully use it, the longer it lasts. The average car doesn’t last a lot more than 150k miles, at least outside of California.
The problem with range anxiety becomes more clear. The 85kwh Tesla lets you do your daily driving around your city with no range anxiety. That’s great. But to get that you buy a huge battery pack. But you only use that extra range rarely, though you spend a lot to get it. Most trips can actually be handled by the 70 mile range Leaf, though with some anxiety. You only need all that extra battery for those occasional longer trips. You spend a lot of extra money just to use the range from time to time. read more »
There’s been a lot of press on robocars in the last few months, and a lot of new writers expressing views. Reading this, I have encountered a recurring set of issues and concerns, so I’ve prepared an article outlining these top myths and explaining why they are not true.
Perhaps of strongest interest will be one of the most frequent statements — that Vehicle to Vehicle (V2V) communication is important, or even essential, to the deployment of robocars. The current V2V (and Vehicle to Infrastructure) efforts, using the DSRC radio spec are quite extensive, and face many challenges, but to the surprise of many, this is largely orthogonal to the issues around robocars.
The cars will need special dedicated roads and lanes
This only works when all cars are robocars and human driving is banned
We need radio links between cars to make this work
We wont see self-driving cars for many decades
It is a long time before this will be legal
How will the police give a robocar a ticket?
People will never trust software to drive their car
They can’t make an OS that doesn’t crash, how can they make a safe car?
We need the car to be able to decide between hitting a schoolbus and going over a cliff
The cars will always go at the speed limit
You may note that this is not my first myths FAQ, as I also have Common objections to Robocars written when this site was built. Only one myth is clearly in both lists, a sign of how public opinion has been changing.
I’m back from CES, and there was certainly a lot of press over two pre-robocar announcements there:
The first was the Toyota/Lexus booth, which was dominated by a research car reminiscent of the sensor-stacked vehicles of the DARPA grand challenges. It featured a Velodyne on top (like almost all the high capability vehicles today) and a very large array of radars, including six looking to the sides. Toyota was quite understated about the vehicle, saying they had low interest in full self-driving, but were doing this in order to research better driver assist and safety systems.
The Lexus booth also featured a car that used ultrasonic sensors to help you when backing out of a blind parking space.
These sensors let you know if there is somebody coming down the lane of the parking lot.
Audi did two demos for the press which I went to see. Audi also emphasized that this is long-term concept stuff, and meant as research work to enhance their “driver in the loop systems.” They are branding these projects “Piloted Parking” and “Piloted Driving” to suggest the idea of an autopilot with a human overseer. However, the parking system is unmanned, and was demonstrated in the lot of the Mandarin Oriental. The demo area was closed off to pedestrians, however.
The parking demo was quite similar to the Junior 3 demo I saw 3 years ago, and no surprise, because Junior 3 was built at the lab which is a collaboration between Stanford and VW/Audi. Junior 3 had a small laser sensor built into it. Instead, the Piloted Parking car had only ultransonic sensors and cameras, and relied on a laser mounted in the parking lot. In this appraoch, the car has a wifi link which it uses to download a parking lot map, as well as commands from its owner, and it also gets data from the laser. Audi produced a mobile app which could command the car to move, on its own, into the lot to find a space, and then back to pick up the owner. The car also had a slick internal display with pop-up screen.
The question of where to put the laser is an interesting one. In this approach, you only park in lots that are pre-approved and prepared for self-parking. Scanning lasers are currently expensive, and if parking is your only
application, then there are a lot more cars then there are parking lots and it might make sense to put the expensive
sensor in the lots. However, if the cars want to have the laser anyway for driving it’s better to have the sensor
in the car. In addition, it’s more likely that car buyers will early adopt than parking lot owners.
In the photo you see the Audi highway demo car sporting the Nevada Autonomous Vehicle testing licence #007. Audi announced they just got this licence, the first car maker to do so. This car offers “Piloted Driving” — the driver must stay alert, while a lane-keeping system steers the car between the lane markers and an automatic cruise control maintains distance from other cars. This is similar to systems announced by Mercedes, Cadillac, VW, Volvo and others. Audi already has announced such a system for traffic jams — the demo car also handled faster traffic.
Audi also announced their use of a new smaller LIDAR sensor. The Velodyne found on the Toyota car and Google cars is a large, roof-mounted device. However, they did not show a car using this sensor.
Audi also had a simulator in their booth showing a future car that can drive in traffic jams, and lets you take a video phone call while it is driving. If you take control of the car, it cuts off the video, but keeps the audio. read more »
Happy 2013: Here are some articles I bookmarked last year that you may find of interest this year.
An NBC video about the AutonoMOUS team in Berlin which is one of the more advanced academic teams, featuring on-street driving, lane-changes and more.
An article about “dual mode transport” which in this case means all sorts of folding bikes and scooters that fit into cars. This is of interest both as competition to robocars (you can park remotely and scoot in, competing with one of the robocar benefits) and interesting if you consider the potential of giving limited self-driving to some of these scooters, so they can deliver themselves to you and you can take one way trips. The robocar world greatly enables the ability to switch modes on different legs of a trip, taking a car on one leg, a bike on another, a subway on a 3rd and a car back home. Now add a scooter for medium length trips.
While I’ve pointed to many videos and sources on the Google car, rather than talk about it myself, if you want a fairly long lecture, check out this talk by Sebastian Thrun at the University of Alberta.
The Freakonomics folks have caught the fever, and ask the same question I have been asking about why urban and transportation planners are blind to this revolution in their analysis
You may have read my short report last year on the Santa Clara Law conference on autonomous vehicles. The Law Review Issue is now out with many of those papers. I found the insurance and liability papers to be of of the most use — so many other articles on those topics miss the boat.
It’s been a while since I’ve done a major new article on long-term consequences of Robocars. For some time I’ve been puzzling over just how our urban spaces will change because of robocars. There are a lot of unanswered questions, and many things could go both ways. I have been calling for urban planners to start researching the consequences of robocars and modifying their own plans based on this.
While we don’t know enough to be sure, there are some possible speculations about potential outcomes. In particular, I am interested in the future of the city and suburb as robocars make having ample parking less and less important. Today, city planners are very interested in high-density development around transit stops, known as “transit oriented development” or TOD I now forecast a different trend I will call ROD, or robocar oriented development.
For a view of how the future of the city might be quite interesting, in contrast to the WALL-E car-dominant vision we often see.
Earlier I wrote an essay on robocar changes affecting urban planning which outlined various changes and posed questions about what they meant. In this new essay, I propose answers for some of those questions. This is a somewhat optimistic essay, but I’m not saying this is a certain outcome by any means.
As always, while I do consult for Google’s project, they don’t pay me enough to be their spokesman. This long-term vision is a result of the external work found on this web site, and should not be taken to imply any plans for that project.
There’s been much debate in the USA about High Speed Rail (HSR) and most notably the giant project aimed at moving 20 to 24 million passengers a year through the California central valley, and in particular from downtown LA to downtown San Francisco in 2 hours 40 minutes.
There’s been big debate about the projected cost ($68B to $99B) and the inability of projected revenues to cover interest on the capital let alone operating costs. The project is beginning with a 130 mile segment in the central valley to make use of federal funds. This could be a “rail to nowhere” connecting no big towns and with no trains on it. By 2028 they plan to finally connect SF and LA.
The debate about the merits of this train is extensive and interesting, but its biggest flaw is that it is rooted in the technology of the past and present day. Indeed, HSR itself is around 50 years old, and the 350 kph top speed of the planned line was attained by the French TGV over 30 years ago.
The reality of the world, however, is that technology is changing very fast, and in some fields like computing at an exponential rate. Transportation has not been used to such rapid rates of change, but that protection is about to end. HSR planners are comparing their systems to other 20th century systems and not planning for what 2030 will actually hold.
At Singularity University, our mission is to study and teach about the effects of these rapidly changing technologies. Here are a few areas where new technology will disrupt the plans of long-term HSR planners:
Cars that can drive and deliver themselves left the pages of science fiction and entered reality in the 2000s thanks to many efforts, including the one at Google. (Disclaimer: I am a consultant to, but not a spokesman for that team.)
Readers of my own blog will know it is one of my key areas of interest.
By 2030 such vehicles are likely to be common, and in fact it’s quite probable they will be able to travel safely on highways at faster speeds than we trust humans to drive. They could also platoon to become more efficient.
Their ability to deliver themselves is both boon and bane to rail transit. They can offer an excellent “last/first mile” solution to take people from their driveways to the train stations — for it is door to door travel time that people care about, not airport-to-airport or downtown-to-downtown. The HSR focus on a competitive downtown-to-downtime time ignores the fact that only a tiny fraction of passengers will want that precise trip.
Self-delivering cars could offer the option of mobility on demand in a hired vehicle that is the right vehicle for the trip — often a light, efficient single passenger vehicle that nobody would buy as their only car today. These cars will offer a more convenient and faster door-to-door travel time on all the modest length trips (100 miles or less) in the central valley. Because the passenger count estimates for the train exceed current air-travel counts in the state, they are counting heavily on winning over those who currently drive cars in the central valley, but they might not win many of them at all.
The cars won’t beat the train on the long haul downtown SF to downtown LA. But they might well be superior or competitive (if they can go 100mph on I-5 or I-99) on the far more common suburb-to-suburb door to door trips. But this will be a private vehicle without a schedule to worry about, a nice desk and screen and all the usual advantages of a private vehicle.
Improved Air Travel
The air travel industry is not going to sit still. The airlines aren’t going to just let their huge business on the California air corridor disappear to the trains the way the HSR authority hopes. These are private companies, and they will cut prices, and innovate, to compete. They will find better solutions to the security nightmare that has taken away their edge, and they’ll produce innovative products we have yet to see. The reality is that good security is possible without requiring people arrive at airports an hour before departure, if we are driven to make it happen. And the trains may not remain immune from the same security needs forever.
On the green front, we already see Boeing’s new generation of carbon fiber planes operating with less fuel. New turboprops are quiet and much more efficient, and there is more to come.
The fast trains and self-driving cars will help the airports. Instead of HSR from downtown SF to downtown LA, why not take that same HSR just to the airport, and clear security while on the train to be dropped off close to the gate. Or imagine a self-driving car that picks you up on the tarmac as you walk off the plane and whisks you directly to your destination. Driven by competition, the airlines will find a way to take advantage of their huge speed advantage in the core part of the journey.
Self-driving cars that whisk people to small airstrips and pick them up at other small airstrips also offer the potential for good door-to-door times on all sorts of routes away from major airports. The flying car may never come, but the seamless transition from car to plane is on the way.
We may also see more radical improvements here. Biofuels may make air travel greener, and lighter weight battery technologies, if they arrive thanks to research for cars, will make the electric airplane possible. Electric aircraft are not just greener — it becomes more practical to have smaller aircraft and do vertical take-off and landing, allowing air travel between any two points, not just airports.
These are just things we can see today. What will the R&D labs of aviation firms come up with when necesessity forces them towards invention?
Rail technology will improve, and in fact already is improving. Even with right-of-way purchased, adaptation of traditional HSR to other rail forms may be difficult. Expensive, maglev trains have only seen some limited deployment, and while also expensive and theoretical, many, including the famous Elon Musk, have proposed enclosed tube trains (evacuated or pneumatic) which could do the trip faster than planes. How modern will the 1980s-era CHSR technology look to 2030s engineers?
Decades after its early false start, video conferencing is going HD and starting to take off. High end video meeting systems are already causing people to skip business trips, and this trend will increase. At high-tech companies like Google and Cisco, people routinely use video conferencing to avoid walking to buildings 10 minutes away.
Telepresence robots, which let a remote person wander around a building, go up to people and act more like they are really there are taking off and make more and more people decide even a 3 hour one-way train trip or plane trip is too much. This isn’t a certainty, but it would also be wrong to bet that many trips that take place today just won’t happen in the future.
Like it or not, in many areas, sprawl is increasing. You can’t legislate it away. While there are arguments on both sides as to how urban densities will change, it is again foolish to bet that sprawl won’t increase in many areas. More sprawl means even less value in downtown-to-downtown rail service, or even in big airports. Urban planners are now realizing that the “polycentric” city which has many “downtowns” is the probable future in California and many other areas.
That Technology Nobody Saw Coming
While it may seem facile to say it, it’s almost assured that some new technology we aren’t even considering today will arise by 2030 which has some big impact on medium distance transportation. How do you plan for the unexpected? The best way is to keep your platform as simple as possible, and delay decisions and implementations where you can. Do as much work with the knowledge of 2030 as you can, and do as little of your planning with the knowledge of 2012 as you can.
That’s the lesson of the internet and the principle known as the “stupid network.” The internet itself is extremely simple and has survived mostly unchanged from the 1980s while it has supported one of history’s greatest whirlwinds of innovation. That’s because of the simple design, which allowed innovation to take place at the edges, by small innovators. Simpler base technologies may seem inferior but are actually superior because they allow decisions and implementations to be delayed to a time when everything can be done faster and smarter. Big projects that don’t plan this way are doomed to failure.
None of these future technologies outlined here are certain to pan out as predicted — but it’s a very bad bet to assume none of them will. California planners and the CHSR authority need to do an analysis of the HSR operating in a world of 2030s technology and sprawl, not today’s.
While there had been many rumous that Mercedes would introduce limited self-driving in the 2013 S-class, that was not to be, however, it seems plans for the 2014 S-class are much more firm. This car will feature “steering assist” which uses stereo cameras and radar to follow lanes and follow cars, along with standard ACC functions. Reportedly it will operate at very high speeds.
There’s also a nice article on the Mercedes test facility. They are well known for their interesting test facilities, and this one uses an inflatable car being towed on a test track, making it safe to hit the car if there is a problem.
Media sources are also reporting that Google (disclaimer, they are a client) has hired Ron Medford, deputy director of the National Highway Transportation Safety Agency, which sets the vehicle safety standards and is currently researching how to certify self-driving cars.
On Tuesday I stopped by the Atlantic’s Big Science day where Chris Gerdes of Stanford’s CARS centre announced results of their race between a robocar and humans on a racetrack. The winner — the humans, but only by a small margin. The CARS team actually studied human driver actions to program their car, but found the human drivers have gotten very good at squeezing most of the available performance out of the vehicles, leaving little room for the robot to improve.
Another result reaffirmed studies of passenger reactions. People taken through the first lap were quite scared, but by the next lap they relaxed and gained confidence in the system. This result shows up time and time again, and has convinced me that while many people tell me they think robocars will not become popular because people will be too scared to ride them, those people are wrong, even about their own behaviour. Most of them, at least.
Also on the Stanford front, Bryant Walker Smith, who has decided to make robocar law a specialty, has released an analysis of the legality of robocars in the USA. The conclusion — robocars which have a human occupant who can take the wheel in the event of a problem are probably legal in almost all states, not just the states that have explicitly made them legal.
DARPA humanoid robot contest includes driving
DARPA ran the 3 grand challenges for robocars but stopped in 2007 after the urban challenge. Their latest challege contest involves making humanoid robots, but the DARPA Robotics Challenge includes a phase where your robot should be able to do a variety of tasks on rough terrain, including getting into a car and driving it. There are 4 tracks to the challenge. 3 are in the physical world, with either provided robots or team-built robots. The 4th is in the virtual world which will allow smaller teams to compete without the cost of working with a physical robot. I have written before about the opportunities of a robocar simulator for testing and contests, and so I am eager to see how this simulator develops.
Research in China has advanced. The National Natural Science Foundation has announced the goal of diong a short drive near Beijing and finally a long trip all the way to Shenzhen, 2400km away. This project primarily uses vision and radar, so it will be interesting to see if they can do this reliably without lasers.
Three big automaker announcements — and not about V2V even though the ITS World Congress is going on this week.
First, Nissan, whose self-parking Leaf I just wrote about has also announced a steer-by-wire system and tests of a car that will swerve to avoid a sudden obstacle. Of course almost all cars have power steering, but in a steer-by-wire car there is no mechanical linkage by default from the steering wheel to the steering motors. This allows a wheel to have “software defined feel” and is good for eventual robocars. In such cars a fail-safe restores a mechanical link if the main system fails.
However, the swerving car, which is demonstrated avoiding a cardboard pedestrian which jumps out into the road, is a new level of technology for major car makers. (Braking for these obstacles has been done for a while.)
Not much later, Jerusalem company MobilEye announced they had converted an Audi A7 to self-drive using 5 of their cameras as well as radar. MobilEye makes the vision system found in a lot of different cars — their specialty is a dedicated chip for vision processing. This article, which is in Hebrew outlines the car, which cost 588K NIS to build.
Volvo, which uses MobilEye, announced today that their 2014 cars would feature a traffic jam assist. Several companies have announced traffic jam assist (which is a low speed lane-keeping plus ACC) but Volvo has put a firm date on it. Also new in Volvo’s system is doing more than following lane markers — it also swerves to follow the car in front of it, as long as that car stays in the lane.
Of course, this does leave open the question of what happens if 2 or more of these start following one another, but that’s some time in the future and they have time to work on it.
In other news, the NHTSA has announced a grant to a team at Virginia Tech to research safety standards for robocar user interfaces. They have in the past stated they think the handoff between manual and automatic is an important safety function they might regulate.
And yes, there is lots of V2V news from the ITS world congress, but my skepticism for most forms of V2V remains high.
Nissan is showing a modified Leaf able to do “valet” park in a controlled parking lot. The leaf downloads a map of the lot, and then, according to Nissan engineers, is able to determine its position in the lot with 4 cameras, then hunt for a spot and go into it. We’ve seen valet park demonstrations before, but calculating position entirely with cameras is somewhat new, mainly because of the issues with how lighting conditions vary. In an indoor parking garage it’s a different story, and camera based localization under the constant lighting should be quite doable.
This other video from Engadget with a more detailed demo shows the view from the car’s cameras, which appear to be on the side mirrors as well as front and back for a synthetic 360 degree view. They also have an Android app for control and the ability to view through the cameras. Alas, chances are low you would get that bandwidth in the parking garage, but it’s a cool demo.
There was a huge raft of press coverage after last week’s signing of the California law. This ranged from polls showing strong acceptance of the tech to editorial critiques about the law being too fanciful or the technology taking jobs. (It is true that there will be job displacement, but at the same time, Americans spend about 50 billion hours driving which is a much larger sink on the GDP.)
Tonight I will be on a panel at the Palo Alto International Film Festival at 5pm. Not on robocars, but on the role of science fiction in movies in changing the world. (In a past life, I published science fiction and am on this panel by virtue of my faculty position at Singularity University.)
Tomorrow (Wed Sep 19) I will give a robocars talk at Dorkbot SF in San Francisco. Dorkbot is a regular gathering of “People doing strange things with electricity” and there will be two other sessions.
Last week, the SARTRE project announced it was concluding after a long period of work on highway platooning. Volvo lead the project which demonstrated platoons on test tracks and on some real roads. They also did a number of worthwhile users studies in simulation.
People have been interested in platooning for a while. The main upsides they are looking for are:
It’s much easier than a Robocar — the platoon is lead by a truck with a professional driver who handles everything with human intelligence
Putting the car at short spacings can result in a huge increase in highway capacity, though you tend to want somewhat larger headways around the convoys
There is fuel saving — about 10% or so for the lead vehicle, and up to 30% for following vehicles, at spacings of about 4 to 6m. This is not quite as much as people hoped but it is real.
The equipment in the following cars is simple — V2V radios and possibly some radar for backup.
Unfortunately, platooning comes with some downsides as well
If you have an accident, it can be catastrophic as you might crash a whole convoy of vehicles.
Non-platoon drivers may interfere with the convoy. The gaps must be kept small enough that nobody tries to enter them. A non-member in the middle of the convoy is bad news. You need small gaps to save fuel too.
Trucks must go only at the front of the convoy due to their longer stopping distance. New trucks must insert in the middle. Cars can insert more easily at the end of the convoy.
Convoys in the right lane can make it harder for people merging, and in general they can present a barrier to traffic.
Driving with a short gap is disconcerting. Behind a truck, you can’t even see the lane markers.
In rain, your windshield gets completely washed out with spray (and sometimes salt spray) which is even more disconcerting.
Following cars get hit by small stones and debris from the forward vehicle. After a long period of following, windshields are unacceptably chipped or cracked.
While radar is the primary means of tracking the car in front, and almost all vehicles do a nice radar reflection from the rear licence plate, many vehicles have other reflections further forward. You must avoid trying to follow 4m behind the front of a truck! To help this, vehicles in the tests had superior radar reflectors mounted on them.
For good workable convoys, some of these problems need to be solved. It could be that in rain convoys must spread out (losing a lot of the fuel saving) though there is the danger of cars cutting in.
Convoys with longer gaps can still increase road capacity a lot, but they probably have to be robocar convoys. Robocar convoys can handle cars trying to cut into the gaps. They may wish to start honking if somebody cuts-in (and the car in front might also flash its rear lights and slow slightly to make it very clear to the cut-in that they should not have done this.) This would be a problem when convoys are new, as people might not know what it all means, though they would have tried to go into a space that is clearly too small to safety enter. Cars in convoys might need to have a screen on the back that can display a sign “You have barged into a convoy, change lanes immediately or be reported to police.”
Robocars could handle the rain to some degree, but even their laser sensors would not like operating in heavy spray, though their radars would get excellent returns from a reflector on the vehicles.
The stone chip problem is harder to solve. Robocars capable of full auto operation could try to protect their windshields, but this is disconcerting to occupants. And the rest of the car gets stone chips too.
It could be that platooning is only practical with vehicles that are dedicated to it, such as highway commute vehicles and long distance highway vehicles. Built for this purpose, they would just accept the stone chips as part of life. They might come with extra heavy duty wipers or other ways to deal with the rain. And they would be full robocars, able to handle disconnects and independent operation.
This result will disappoint those who felt platoons were a good early technology. I have felt they also suffered from a critical mass problem. To use a platoon, you would need to find one, and until the density of lead vehicles was high enough, you might not find one. You could do it at rush hour with mobile apps that track the presence of lead vehicles so you can time your departure to find one — you might even have an appointment for every commute. And they might run only on nice clean highways on dry days and still be valuable. But less valuable, I am afraid.
On lower speed roads the fuel saving is not much, but the problems are less. There are traffic lights on most low speed roads though which present another problem.
A round-up of just some of the recent robocar news:
Stanford Shelly at 120mph
While the trip up Pikes Peak by Stanford’s Audi TT did not offer the high speeds we had hoped for, they have recently being doing some real race driving tests, clocking the car around a track at 120mph. Even more impressive because this car drives with limited sensors. Here the goal is to test computer driven high-speed tactics — rounding corners, climbing hills and more. While they didn’t quite reach the times of professional drivers, chances are someday they will, just from the perfect understanding of physics.
Driving this fast is hard in the real world because you’re going beyond the range of most sensors (radar and special lidars can go further, and cameras can see very far but are not reliable in all lighting.) The Stanford team had a closed track to they were able to focus on cornering and skidding.
KPMG report on self-driving cars
The consulting firm KPMG has released an extensive report on self-driving cars. While it doesn’t contain too much that is new to readers of this site and blog, it joins the group which believes that car-to-car communication is going to be necessary for proper deployment of robocars. I don’t think so, and in fact think the idea of waiting for it is dangerous.
Speaking of V2V communication
For some time the V2V developers have been planning a testbed project in Ann Arbor, MI. They’ve equipped 3000 cars with “here I am” transponders that will broadcast their GPS data (position and velocity) along with other car data (brake application, turn signals, etc.) using DSRC. It is hoped that while these 3000 civilian cars will mostly wander around town, there will be times when the density of them gets high enough that some experiments on the success of DSRC can be made. Most of the drivers of the cars work in the same zone, making that possible.
If they don’t prove the technology, they probably won’t get the hoped-for 2013 mandate that all future cars have this technology in them. If they don’t get that, the 75mhz of coveted spectrum allocated to DSRC will get other hungry forces going after it.
I owe readers a deeper analysis of the issues around vehicle-to-vehicle communications.
Google cars clock 300,000 miles
Google announced that our team (they are a consulting client) has now logged 300,000 miles of self-driving, with no accidents caused by the software. It was also acknowledged that the team has also converted a hybrid Lexus RX-450h in addition to the Toyota Prius. Certainly a more
comfortable ride and the new system has very nice looks.
Google will also begin internal testing with team members doing solo commutes in the vehicles. Prior policy is vehicles are always operated off-campus with two staff onboard, as is appropriate in prototype systems.
Political attack ad goes after robocars
Jeff Brandes pushed Florida’s legislation to allow robocar testing and operations in that state, 2nd after Nevada. Now his political opponents have produced an ad which suggests robocars are dangerous and you shouldn’t vote for Mr. Brandes because of his support of them. While we should expect just about anything in attack ads, this is a harbinger of the real debate to come. I doubt the authors of the ads really care about robocars — they just hope to find anything that might scare voters. My personal view, as I have said many times, is that while the technology does have to go through a period where it is less safe because it is being prototyped and developed, the hard truth is that the longer we wait to deploy the technology, the more years we rack up with 34,000 killed on the roads in the USA and 1.2 million worldwide. And Florida’s seniors are among the first on the list to need robocars. Is Jim Frishe’s campaign thinking about that?
Collision Warning strongly pushed in Europe
The EU is considering changing its crash-safety rules so that a car can’t get a 5-star rating unless it has forward collision warning, or even forward-collision mitigation (where the system brakes if you don’t.) These systems are already proving themselves, with data suggesting 15% to 25% reductions in crashes — which is pretty huge. While the law would not force vendors to install this, there are certain car lines where a 5-star rating is considered essential to sales.
I’m here in Newport beach at the Transportation Research Board’s conference on self-driving vehicles. Today in a pre-session there was discussion of pre-robocar technologies and in particular applications of “managed lanes” and what the might mean for these technologies. Managed lanes are things like HOV/carpool lanes, HOT (carpool+toll), reversible lanes etc. Many people imagine these lanes would be used with pre-robocar technologies like convoys, super-cruise, cooperative ACC, Bus Rapid Transit etc.
As I’ve said before the first rule of robocars is “you don’t change the infrastructure.” First you must make the vehicles operate fully on the existing infrastructure. And people are doing that. But we can also investigate what happens next.
Robocars as many envision them do not thus need dedicated lanes, even though some of the simpler technologies might. Earlier we talked about electrification which is a pretty expensive adaptation. Let’s talk about high speed lanes.
Robocars (or any car) would be of much greater interest to people if they could go very fast in them. On one hand, the ability to work, read, watch video and possibly sleep in a robocar will mean to some that trip time is less important than comfort, and they might actually be happy with a slower trip with fewer disturbances. But sometimes a faster trip is very important, particularly on the long haul.
Today people are working hard to make robocars safe. Eventually they should be able to make them safe even at higher speeds, particularly on freeways that were designed for fairly high speeds. Even human drivers routinely see over 100mph on the autobahns of Germany. Problem is if you want to go 120mph outside of Germany , there’s no road you can easily do that on. The other cars, going 65 to 80mph in the fast lane will get in the way, creating an uncomfortable ride and possibly dangerous situations.
Many of today’s “managed lanes” are primarily for use in rush hour, from 5am to 9am and 3pm to 7pm. In other hours, traffic is very light. What if that special lane does not just become an ordinary lane after rush hour, but instead is converted to another special purpose. There are a lot of different technologies that might be able to become viable with such a lane.
The most interesting one to me is high speed. If the carpool lane switched to being the high-speed-car lane at 9:30am, I actually think a lot of people might very well delay their commutes and shift their hours. A one-hour commute at 8am or a 15 minute trip at 9:30am — not a hard choice for many. And lots of people travel mid-day for various purposes.
The high-speed lane would actually mandate a minimum speed, perhaps 100mph when the road is clear. To get in this lane you would need a car that is certified safe at that speed or above. This might be a robocar, but it might also be a human-driven car with sufficient driver-assist technologies to certify it safe at that speed. The lane would probably only be open in good weather, and would probably revert to ordinary status in the event the main road got congested for whatever reason. Vehicles in the lane would have to be connected vehicles, ready to receive signals about changes to the dynamic status of the lane.
There probably would also be a requirement for efficient vehicles. Wind drag at 120mph costs 4 times as much fuel per mile as wind drag at 60mph. These cars would have to be highly aerodynamic designs. They might also be capable of platooning to further reduce drag, though you would want to wait a while to assure safety before platooning at 120mph. You might insist on alternate fuels or even that they be electric vehicles or other low emission vehicles. It doesn’t matter — I think there are a lot of people who would pay a lot of money to be able to go 120mph.
The lanes in general would need to be separated from the main lanes. Most carpool lanes are already like that, though most of the ones in the SF Bay Area are not this style. Ideally they would be the style that even has a special merge lane at the points where entry and egress from the main lanes are possible.
If such a program were a success we could see more. For example, one could imagine adding an extra lane to Interstate 5 in the California central valley and have it be a high speed lane most of the time. The planned California High Speed Rail, which probably will never be finished, is forecast to cost $68 Billion. 2 extra lanes on I-5 in the central valley south of Sacramento would cost well under a billion, and offer fairly high speed travel to those in the valley — faster door to door than the HSR. And my calculations even suggest that aerodynamic electric vehicles would use less energy per passenger-mile than the HSR. (Definitely if they are shared by as few as 2-3 people or when designed for a platoon.) These teardrop-shaped cars would also be much more efficient than today’s cars when they slow down and ply the ordinary highways and streets.
It is not trivial to go 120mph in a robocar though. Your sensors must be long range so you can stop if they see something. If you want to build infrastructure, here is where the road might have sensors which can report on road obstacles and other vehicles to assure safety. If you’re building a whole high speed lane this is not an issue. The first rule of robocars is written to avoid needing new infrastructure to do ordinary driving and get most places — not to prevent you from taking advantage of new spending that justifies itself.
An MIT team has been working on a car that is “hard to crash.” Called the intelligent co-pilot it is not a self-driving car, but rather a collection of similar systems designed to detect if you are about to hit something and try to avoid it. To some extent, it actually wrests control from the driver.
When I first puzzled over the roadmap to robocars I proposed this might be one of the intermediary steps. In particular, I imagined a car where, in a danger situation, the safest thing to do is to let go of the wheel and have the car get you to a safe state. This car goes further, actually resisting you if you try to drive the car off the road or towards an obstacle.
This is a controversial step, and the reasons are understood by the MIT team. First of all, from a legal liability standpoint, vendors are afraid of overriding the human. If a person is in control of a vehicle and makes a mistake, they are liable. If a machine takes over and saves the day, it’s great, but if the machine takes over and there is an accident — an accident the human could have avoided — there could be high risks to the maker of the machine as well as the occupant. In most designs, the system is set up so that the human has the opportunity for control at all times.
Actually, it’s even worse. A number of car makers are building freeway autopilots which still require attention from the driver in case the lane markers disappear or other problems ensue. One way some of them have built this is to require the driver to touch the wheel every so often to show they are alert. They will beep if the driver does not touch the wheel, and they will even disengage if the driver waits for too long after the beep. Consider what the companies have interpreted the liability system to require: That the right course of action, when the system is driving and the driver has her hands off the wheel, is to disengage and let the vehicle wander freely and possibly careen off the road! Of course, they don’t want the vehicle to do that, but they want to make it clear to the driver that they can’t depend on the system, can’t decide to type a long E-mail while it is running.
And this relates to the final problem of human accommodation. When a system makes people safer, they compensate by being more reckless. For example, anti-lock brakes are great and prevent wheel lock-up on slippery roads — but they cause drivers to feel they have invincible brakes and studies show they drive more aggressively because of them. Only a safe robocar avoids this problem; its decisions will always be based on a colder analysis of the situation.
A hard-to-crash car is still a very good idea. Before a full robocar is available, it can make a lot of sense, particularly for aging people and teens with new licences. But it may never come to market due to liability concerns.
There have been experiments with dedicated lanes in the past, including a special automated lane back in the 90s in San Diego. The problem is much easier to solve (close to trivial by today’s standards) if you have a dedicated lane, but this violates the first rule of robocars in my book — don’t change the infrastructure.
Aside from the huge cost of building the dedicated lanes, once you have built a lane you now have a car which can only drive itself in that dedicated lane. That’s a lot less valuable to the customer, effecitvely you are only get customers who happen to commute on that particular route, rather than being attractive to everybody. And you can’t self-drive on the way to or from the highway, so it is not clear what they mean when they say the driver sets a destination, other than perhaps the planned exit.
Yes, the car is a lot cheaper but this is a false economy. Robocar sensors are very expensive today but Moore’s law and volume will make them cheaper and cheaper over time. Highway lanes are not on any Moore’s law curve, in fact they are getting more expensive with time. And if the lane is dedicated, that has a number of advantages, though it comes with a huge cost.
Of course, today, nobody has a robocar safe enough to sell to consumers for public streets. But I think that by the early 2020s, when this study might recommend completing a highway, the engineers would open up the new lane and find that while it’s attractive for its regular nature and especially attractive if it is restricted and thus has lighter and more regular traffic, the cars are already able to drive on the regular lanes just fine.
A better proposal, once robocars start to grow in popularity, would be to open robocar lanes during rush hour, like carpool lanes. These lanes would not be anything special, though they would feature a few things to make the car’s job easier, such as well maintained markings, magnets in the road if desired, no changes in signage or construction without advance notice etc. But most of all they would be restricted during rush hour so that cars could take advantage of the smooth flow and predictable times that would come with all cars being self-driving. Unless humans kept taking over the cars and braking when they got scared or wanted to look at an accident in the other lanes, these lanes would be metered and remain free of traffic jams. However, you need enough robocar flow to justify them since if you only use half the capacity of a lane it is wasteful. On the other hand, such lanes could be driven by the more common “super cruise” style cars that just do lane following and ACC.
Hats off to the video embedded below, which was prepared for a futuristic transportation expo in my home town of Toronto.
Called the PAT (People and Things) this video outlines the UI and shows a period in the day of a robotic taxi/delivery vehicle as it moves around Toronto picking up people and packages.
I first learned about the video from a new blog on the subject of consumer self driving cars — as far as I know the second serious one to exist after this one. The Driverless Car HQ started up earlier this year and posts with a pretty solid volume. They are more comprehensive in posting various items that appear in the media than I am, and cover some areas I don’t, so you may want to check them out. (That’s a conscious choice on my part, as I tend not to post links to stories that I judge don’t tell us much new. An example would be that the SARTRE road train just did a demo in Spain last month, but it was not much different from demos they had done before.)
Of course, as I said earlier, sadly “Driverless Car” is one of my least favourite terms for this technology, but that doesn’t impede the quality of the blog. In addition, while I do report news on the Google car on this blog, I tend to refrain from commentary due to being on that team, and the folks at DCHQ are not constrained this way.
Face recognition of passengers as they approach the car
Automatic playing of media for the passengers (apparently resuming from media paused earlier in some cases)
Doing package delivery work when needed
Self-cleaning after each passenger
Optional ride-share with friends
In-car video conferencing on the car’s screens
Offering the menu of a cafe which is the destination of a trip. (Some suspect this is a location-based ad spam, but I think it’s a more benign feature because the passenger is picking up his ride-share friend at the cafe.)
And the UIs are slick, if a bit busy, and nicely done.
The concept vehicle at the Brickworks is fairly simple but does present some ideas I have felt are worthwhile, such as single passenger vehicles, face to face seating etc. It’s a bit too futuristic, and not aerodynamic. In the concept, it adjusts for the handicapped. I actually think that’s the reverse of what is likely to happen. Rather than making all cars able to meet all needs, it makes more sense to me to have specialized cars that are cheaper and more cost effective at their particular task, and have dedicated (more expensive) vehicles for wheelchairs. (For example, I like the hollow vehicles like the Kenguru.) I think you serve the disabled better for the same money by having these specialized vehicles — the wait may be slightly longer, but the vehicle can be much better at serving the individual’s needs.
Ford, which has already touted the value of robocars, has announced plans to do a traffic-assist autopilot system sometime mid-decade. Ford joins Mercedes, VW/Audi and Cadillac in announcing such systems. Ford’s vehicle will also offer automatic parking in perpendicular parking spots. For some time many cars have offered automated parallel parking. Since most people do not find perpendicular parking all that difficult, perhaps their goal here is very tight spaces (though that would require getting out of the car and blocking the rude driver, which I have found out only gets your car vandalized) or possibly parking in a personal garage that is very thin.
AUVSI and Mercedes
On the negative front, Mercedes appears to have backed off their plan to offer a traffic jam assistant in the 2013 S class. Earlier in June I attended the AUVSI “Driverless Car Summit” in Detroit, and Mercedes indicated that while they do have that technology in their F.800 concept car, this is only a prototype. As currently set up, the Mercedes system requires you to touch the wheel every 8 seconds. Honda was promoting this in 2006. Mercedes also showed their “6D” stereo vision based system which demonstrated impressive object tracking. They also claimed it does as well in differing light conditions, which would be a major breakthrough.
Some other notes from the conference:
There was effectively universal hate for the term “driverless car.” I join the haters, since the car has a driver, but it’s a computer. No other term won big support, though.
While AUVSI is about unmanned military vehicles, they put on a nicely demilitarized conference, which was good.
There were still a lot of fans of DSRC (a car data radio protocol) and V2V communications. Some from that community have now realized they went down the wrong path but a lot had made major career investments and will continue to push it, including inside the government.
The NHTSA is doing a research project on how they might regulate safety standards. They have not laid out a strategy but will be looking at sensor quality, low level control system squality, UI for the handoff between manual and self-driving and testing methodology.
I liked Mercedes’ terms for various modes of self-driving: Feet off, Hands off, Eyes off and Body out. The car companies are aiming at hands off, Google is working on Eyes Off but Body out (which means being so good that the car can operate without anybody in it or without any attention from the occupant) is the true robocar and the long term goal for many but not all projects.
Continental showed more about their own cruising system that combines lane-keeping and automatic cruise-control. They now say they have the 10,000 miles of on-road testing needed for the Nevada testing licence, but have not yet decided if they will get one. There is some question is what they are doing requires a licence under the Nevada regulations. (I suspect it does not.) However, they were quizzed as to whether they were testing in Nevada without a licence, which they deny. Continental says their system is built entirely from parts that will be “production parts” as of early 2013.
Legal and states panels showed progress but not too much news. States seem to be pleased so far.
The National Federation for the Blind showed off their blind driving challenge. They have become keen on building a car which has enough automation for a blind person to operate but still uses the blind driver’s skills (such as hearing and thinking) to make the task possible. This is an interesting goal for the feeling of autonomy, but I suspect it is more likely they will just get full-auto cars sooner, and they accept this is likely.
Know me by my flyer number and don’t repeat things to me I’ve already certified as knowing, like safety rules
Know my language (I input it, after all) and don’t bother me with announcements other than in my best understood language
Show me most things as text, perhaps in a crawl under my show. If need be, have me confirm I understand.
Tailor the message to my age and my location in the plane. Show me exits on the screen for my seat.
Cut back on the spam about how great your airline is, how wonderful the FF plan is or why I should buy duty free.
Today, instead you can see the visible frustration on the faces of flyers as their movie is interrupted so that they can here the translation into Russian of the long announcement they just heard in German and English.
Having good custom in-flight entertainment is good, and considered a major competitive feature, but already I see more and more people preferring to put out a tablet, even when they have a super-fancy system in the first-class seats. The tablet of course does not have the interruptions (even for the tiny number of real announcements such as in our case last week, “we can’t get the landing gear doors to close so we’re dumping fuel and returning to the airport”) and it also has, if you prepare it, customized entertainment that you know you want to watch.
Frankly, I am not sure who programs the video selections on many of the airlines but I have to suspect they don’t just try to get the best movies with good reviews. They either try to get the cheapest movies or have deals with certain studios — it’s amazing how few quality films they might have in a selection of 100 movies.
I also remain disappointed at how badly implemented most of the in-flight systems I have seen. They are all slow, with highly noticeable lags after keypresses, poor touchscreens, freezes and crashes. Any tablet or phone puts them to shame when it comes to UI and responsiveness. And to top it off, they are huge — on main airlines, many of the seats have reduced footroom to fit the box for the video system. (It also has other in-seat electronics I presume but still, it’s about 10x bigger than it needs to be.) This is odd particularly since in planes floor space and weight are at such a premium. A tablet computer, either fixed in the seat or on an anti-theft power/data tether, would provide a better system — smaller, lighter, better UI, cheaper, better screen — in just about every way. Of course when they first designed these seats years ago they did not have cheap tablets but there is little excuse to continue installing the old ones.
Wait, how could they have known? How could they have not known. It’s 2012. We’ve known for decades now that each year computer products get smaller, faster, cheaper and superior in major ways. When you are designing a system to install in the future, it’s a mistake to design it based on the current technology. You should bet that something better will be along and make your design adaptable to it. If nothing else your standard design is going to get faster and higher resolution — which makes the slow response time of the existing systems inexplicable.
Many airlines are starting to offer satellite TV. That’s better than the old limited selections (or in particular a single bad movie) but actually not too appealing. Aside from being full of commercials and ignoring your schedule, with TV the announcements and interruptions make you miss crucial parts of your show as they talk over them. More than once I’ve been watching a show on an airline to have them talk over the climax of the film.
I’m whining a lot but it’s because I do believe this is important. Truth is that on a flight you are often tired and cramped, and reading and working are not tremendously comfortable. I bring a book but read at a reduced speed. Having nice noise-cancelling headphones and a good in-flight entertainment system with quality content can make a make a flight much better, and it’s a shame that so many things are obviously wrong with the systems they have built. Today’s flights are stressful in any cabin, and a quiet and uninterrupted experience would do a lot to increase customer satisfaction.