Last week, Volvo was demoing some new collision avoidance features in their S60. I’ve talked about the S60 before, as it surprised me putting pedestrian detection into a car before I expected it to happen. Unfortunately in an extreme case of demo disease known to all computer people, somebody has made an error with the battery, and in front of a crowd of press, the car smashed into the truck it was supposed to avoid. The wired article links to a video.
Poor Volvo, having this happen in front of all the press. Of course, their system is meant to be used in human driven cars, warning the driver and braking if the driver fails to act — not in a self-driving vehicle. And they say that had their been a driver there would have been an indication that the system was not operating.
While this mistake is the result of a lack of maturity in the technology, it is important to realize that as robocars are developed there will be crashes, and some of the crashes will hurt people and a few will quite probably kill people. It’s a mistake to assume this won’t happen, or not to plan for it. The public can be very harsh. Toyota’s problems with their car controllers (if that’s where the problems are — Toyota claims they are not — have been a subject of ridicule for what was (and probably still is) one of the world’s most respected brands. The public asks, if programmers can’t program simple parts of today’s cars, can they program one that does all the driving?
There are two answers to that. First of all, they can and do program computerized parts of today’s cars all the time and by and large have perfect safety records.
But secondly, no they can’t make a complete driving system perfectly safe, certainly not at first. It is a complex problem and we’ll wait a long time before the accident rate is zero. And while we wait, human drivers will kill millions.
Our modern society has always had a tough time with that trade-off. Of late we’ve been coming to demand perfect safety, though it is impossible. Few new products are allowed out if it is known that they will have any death rate due to their own flaws. Even if those flaws are not known in the specific, but are known to be highly likely to exist in some fashion. American juries, faced with minutes of a meeting where the company decided to “release the product, even though predictions show that bugs will kill X people” will punish the company nastily, even though the alternative was “don’t release and have human drivers kill 10X people.” The 9X who were saved will not be in the courtroom. This is one reason robocars may arise outside the USA first.
Of course, there might be cases the other way. A drunk who kills somebody when he could have taken a robocar might get a stiffer punishment. A corporation that had its employees drive when robotic systems were clearly superior might find a nasty judgement — but that would require that it was OK to have the cars on the road in the first place.
But however this plays out, developers must expect there will be bugs, an bugs with dire consequences. Nobody will want those bugs, and all the injuries will be tragic, but so is being too cautious on deployment. Can the USA figure a way to make that happen?
This weekend I attended the annual “Robogames” competition, which took place here in the Bay Area. Robogames is mostly a robot battle competition, with a focus on heavily armed radio-controlled robots fighting in a protected arena. For several years robot fighting was big enough to rate some cable TV shows dedicated to it. The fighting is a lot of fun, but almost entirely devoid of automation — in fact efforts to use automation in battle robots have mostly been a failure.
The RC battles are fierce and violent, and today one of the weapons of choice is something heavy that spins at very high speed so that it builds up a lot of angular momentum and kinetic energy, to transfer into the enemy. People like to see robots flying through the air and losing parts to flying sparks. (I suspect this need to make robots very robust against attack makes putting sensors on the robots for automation difficult, as many weapons would quickly destroy a lot of popular sensors types.)
The games also featured a limited amount of automated robot competition. This included some lightweight (3lb and 1lb) automated battles which I did not get to watch, and some some hobby robot competitions for maze-running, line following, ribbon climbing and LEGO mindstorms. There was also semi-autonomous robot battle called “kung fu” where humanoid robots who take high level commands (like punch, and step) try to push one another over. There is also sumo, a game where robots must push the other robot out of the ring.
I had hoped the highlight would be the Robo-magellan contest. This is a hobbyist robot car competition, usually done with small robots 1 to 2 feet in length. Because it is hobbyists, and often students, the budgets are very small, and the contest is very simple. Robots must make it through a simple outdoor course to touch an orange cone about 100 yards away. They want to do this in the shortest time, but for extra points they can touch bonus cones along the way. Contestants are given GPS coordinates for the target cones. They get three tries. In this particular contest, to make it even easier, contestants were allowed to walk the course and create some extra GPS waypoints for their robots.
These extra waypoints should have made it possible to do the job with just a GPS and camera, but the hobbyists in this competition were mostly novices, and no robot reached the final cone. The winner got within 40 feet on their last run, but no performance was even remotely impressive. This was unlike past years, where I was told that 6 or more robots would reach the target and there would be real competition. This year’s poor showing was blamed on budgets, and the fact that old teams who had done well had moved on from the sport. Only 5 teams showed up.
The robots were poor for sensors. While all would have a GPS, in 1 or 2 cases the GPS systems failed and the robots quickly wandered into things. A few had sonar or touch-bars for obstacle detection, but others did not, and none of them did their obstacle detection well at all. For most, if they ran into something, that was it for that race. Some used a compass or accelerometers to help judge when to turn and where to aim, since a GPS is not very good as a compass. read more »
I’ve been predicting a great deal of innovation in cars with the arrival of robocars and other automatic driving technologies. But there’s a lot of other computerization and new electronics that will be making its way into cars, and to make that happen, we need to make the car into a platform for innovation, rather than something bought as a walled garden from the car vendor.
In the old days, it was fairly common to get a car without a radio, and to buy the radio of your choice. This happened even in higher end cars. However, the advantages in sound quality and dash integration from a factory-installed radio started to win out, especially with horizontal market Japanese companies who were both good at cars and good at radios.
For real innovation, you want a platform, where aftermarket companies come in and compete. And you want early adopters to be able to replace what they buy whenever they get the whim. We replace our computers and phones far more frequently than our cars and the radios inside them.
To facilitate this, I think the car’s radio and “occupant computer” should be merged, but split into three parts:
The speakers and power amplifier, which will probably last the life of the car, and be driven with some standard interface such as 7.1 digital audio over optical fiber.
The “guts” which probably live in the trunk or somewhere else not space constrained, and connect to the other parts
The “interface” which consists of the dashboard panel and screen, with controls, and any other controls and screens, all wired with a network to the guts.
Ideally the hookup between the interface and the guts is a standardized protocol. I think USB 3.0 can handle it and has the bandwidth to display screens on the dashboard, and on the back of the headrests for rear passenger video. Though if you want to imagine an HDTV for the passengers, its possible that we would add a video protocol (like HDMI) to the USB. But otherwise USB is general enough for everything else that will connect to the guts. USB’s main flaw is its master-slave approach, which means the guts needs to be both a master, for control of various things in the car, and a slave, for when you want to plug your laptop into the car and control elements in the car — and the radio itself.
Of course there should be USB jacks scattered around the car to plug in devices like phones and memory sticks and music players, as well as to power devices up on the dash, down in the armrests, in the trunk, under the hood, at the mirror and right behind the grille.
Finally there need to be some antenna wires. That’s harder to standardize but you can be we need antennas for AM/FM/TV, satellite radio, GPS, cellular bands, and various 802.11 protocols including the new 802.11p. In some cases, however, the right solution is just to run USB 3.0 to places an antenna might go, and then have a receiver or tranceiver with integrated antenna which mounts there. A more general solution is best.
This architecture lets us replace things with the newest and latest stuff, and lets us support new radio protocols which appear. It lets us replace the guts if we have to, and replace the interface panels, or customize them readily to particular cars. read more »
A couple of weeks ago I wrote about the need for a good robocar driving simulator. Others have been on the track even earlier and are arranging a pair of robotic driving contests in simulator for some upcoming AI conferences.
The main contest is a conventional car race. It will be done in the TORCS simulator I spoke of, where people have been building robot algorithms to control race cars for some time, though not usually academic AI researchers. In addition, they’re adding a demolition derby which should be a lot of fun, though not exactly the code you want to write for safety.
This is, however, not the simulator contest I wrote about. The robots people write for use in computer racing simulators are given a pre-distilled view of the world. They learn exactly where the vehicle is, where the road edges are and where other cars are, without error. Their only concern is to drive based on the road and the physics of their vehicle and the track, and not hit things — or in the case of the derby, to deliberately hit things.
The TORCS engine is a good one, but is currently wired to do only an oval racetrack, and the maintainers, I am told, are not interested in having it support more complex street patterns.
While simulation in an environment where all the sensing problems are solved is a good start, a true robocar simulation needs simulated sensors — cameras, LIDAR, radar, GPS and the works — and then software that takes that and tries to turn it into a map of where the road is and where the vehicles and other things are. Navigation is also an important thing to work out. I will try to attend the Portland version of this conference to see this contest, however, as it should be good fun and generate interest.
It is no coincidence that two friends of mine have both founded companies recently to build telepresence robots. These are easy to drive remote control robots which have a camera and screen at head height. You can inhabit the robot, and drive it around a flat area and talk to people by videoconferencing. You can join meetings, go visit people or inspect a factory. Companies building these robots, initially at high prices, intend to sell them both to executives who want to remotely tour remote offices and to companies who want to give cheaper remote employees a more physical presence back at HQ.
There are also a few super-cheap telepresence robots, such as the Spykee, which runs Skype video conferencing and can be had for as low as $150. It’s not very good, and the camera is very low down, and there’s no screen, but it shows just how cheap such a product can get.
“Anybots” QA telepresence robot
When they get down to a price like that, it seems inevitable to me that we will see an emergency services robot on every block, primarily for use by the police. When there is a police, fire or ambulance call to an address, an officer could immediately connect to the robot on that block and drive it to the scene, to be telepresent. The robot would live in a small, powered protective closet either paid for by the city, but more likely just donated by some neighbour on the block who wants the fastest possible emergency response. Called into action, the robot’s garage door would open and the robot would drive out, and probably be at the location of the emergency within 60 to 120 seconds, depending on how densely they are placed. In the meantime actual first responders might also be on the way.
Watching and managing children is one of the major occupations of the human race. A true robot babysitter is still some time in the future, and getting robocars to the level that we will trust them as safe to carry children is also somewhat in the future, but it will still happen much sooner.
Today I want to explore the implications of a robocar that is ready to safely carry children of certain age ranges. This may be far away because people are of course highly protective of their children. They might trust a friend to drive a child, even though human driving records are poor, because the driver is putting her life on the line just as much as the child’s, while the robot is just programmed to be safe, with no specific self-interest.
A child’s robocar can be designed to higher safety standards than an adult’s, with airbags in all directions, crumple zones designed for a single occupant in the center and the child in a 5-point seatbelt. As you know, with today’s modern safety systems, racecar drivers routinely walk away from crashes at 150mph. Making a car that won’t hurt the child in a 40mph crash is certainly doable, though not without expense. A robocar’s ability to anticipate an accident might even allow it to swivel the seat around so that the child’s back is to the accident, something even better than an airbag.
The big issue is supervision of smaller children. It’s hard to say what age ranges of children people might want to send via robocar. In some ways infants are easiest, as you just strap them in and they don’t do much. All small children today are strapped in solidly, and younger ones are in a rear facing seat where they don’t even see the parent. (This is now recommended as safest up to age 4 but few parents do that.) Children need some supervision, though real problems for a strapped in child are rare. Of course, beyond a certain age, the children will be fully capable of riding with minimal supervision, and by 10-12, no direct supervision (but ability to call upon an adult at any time.) read more »
One of the things that’s harder to predict about robocars is what they will mean for how cities are designed and how they evolve. We’re notoriously bad at predicting such things, but it is still tempting.
A world of robocars offers the potential for something I am dubbing the “poor man’s teleporter.” That’s a fleet of comfortable robotaxis that are, while you are in them, a fully functional working or relaxing environment. Such robotaxis would have a desk and large screen and very high speed wireless net connection. They have a comfy reclining chair (or bed) and anything else you need from the office environment. (Keyboards and mice are problematic, as I have discussed elsewhere, but there may be ways to solve that.)
The robotaxi will deliberately pick the most comfortable route for a trip, with few turns, few stops and gentle acceleration. It will gimbal in corners and have an active suspension system eliminating bumps. The moment you enter it, your desktop could appear on the screen, copied from the desk you left (thanks to communication with one of your wearable devices, probably.) You can do high quality videoconferencing, work on the net, or just watch a video or read a book — the enclosed book reader could be set to the page you were last reading elsewhere. If you work in a building with a lobby, the electric robotaxi could enter the lobby and meet you right at the elevator. It might even go vertical and ride up the elevator to get you during less busy times. (For some real science fiction, the robotaxis in Minority Report somehow climbed the buildings and parked in people’s homes.)
For many it would be as though they had not left their desks. Almost all the trip will be productive time. As such, while people won’t want to spend forever in the car, many might find distance and trip time to not be particularly important, at least not for trips around town during the workday. While everybody wants to get home to family sooner, even commute times could become productive times with employers who let the employee treat the travel time as work time. Work would begin the moment you stepped into the car in the morning.
We’ve seen a taste of this in Silicon Valley, as several companies like Google and Yahoo run a series of commute vans for their employees. These vans have nice chairs, spaces for laptops and wireless connectivity into the corporate network. Many people take advantage of these vans and live in places like San Francisco, which may be an hour-long trip to the office. The companies pay for the van because the employees start the workday when they get on it.
This concept will continue to expand, and I predict it will expand into robocars. The question is, what does it mean to how we live if we eliminate the time-cost of distance from many trips? What if we started viewing our robotaxis as almost like a teleporter, something that takes almost no time to get us where we want to go? It’s not really no-time, of course, and if you have to make a meeting you still have to leave in time to get there. It might be easier for some to view typical 15 minute trips around a tight urban area as no-time while viewing 30-60 minute trips as productive but “different time.”
Will this make us want to sprawl even more, with distance not being important? Or will we want to live closer, so that the trips are more akin to teleportation by being productive, short and highly predictable in duration? It seems likely that if we somehow had a real Star-Trek style transporter, we might all live in country homes and transport on demand to where the action is. That’s not coming, but the no-lost-time ride is. We might not be able to afford a house on the nice-walkable-shops-and-restaurants street, but we might live 2 miles from it and always be able to get to it, with no parking hassle, in 4 minutes of productive time.
What will the concept of a downtown mean in such a world? “Destination” retailers and services, like a movie house, might decide they have no real reason to be in a downtown when everybody is coming by robotaxi. Specialty providers will also see no need to pay a premium to be in a downtown. Right now they don’t get walk-by traffic, but they do like to be convenient to the customers who seek them out. Stores that do depend on walk-by traffic (notably cafes and many restaurants) will want to be in places of concentration and walking.
But what about big corporate offices that occupy the towers of our cities? They go there for prestige, and sometimes to make it easy to have meetings with other downtown companies. They like having lots of services for their employees and for the business. They like being near transit hubs to bring in those employees who like transit. What happens when many of these needs go away?
For many people, the choice of where to live is overwhelmingly dominated by their children — getting them nice, safe neighbourhoods to play in, and getting them to the most desired schools. If children can go to schools anywhere in a robocar, how does that alter the equation? Will people all want bigger yards in which to cacoon their children, relying on the robocar to take the children to play-dates and supervised parks? Might they create a world where the child goes into the garage, gets in the robocar and tells it to go to Billy’s house, and it deposits the child in that garage, never having been outside — again like a teleporter to the parents? Could this mean a more serious divorce between community and geography?
While all this is going on, we’re also going to see big strides in videoconferencing and virtual reality, both for adults, and as play-spaces for adults and children. In many cases people will be interacting through a different sort of poor man’s teleporter, this one taking zero time but not offering physical contact.
Clearly, not all of these changes match our values today. But what steps that make sense could we actually take to promote our values? It doesn’t seem possible to ban the behaviours discussed above, or even to bend them much. What do you think the brave new city will look like?
It is often said that cars caused the suburbanization of cities. However, people didn’t decide they wanted a car lifestyle and thus move where they could drive more. They sought bigger lots and yards, and larger detached houses. They sought quieter streets. While it’s not inherent to suburbs, they also sought better schools for kids and safer neighbourhoods. They gave up having nearby shops and restaurants and people to get those things, and accepted the (fairly high) cost of the car as part of the price. Most often for the kids. Childless and young people like urban life; the flight to the suburbs was led by the parents.
This doesn’t mean they stopped liking the aspects of the “livable city.” Having stuff close to you. Having your friends close to you. Having pleasant and lively spaces to wander, and in which you regularly see your friends and meet other people. Walking areas with interesting shops and restaurants and escape from the hassles of parking and traffic. They just liked the other aspects of sprawl more.
They tried to duplicate these livable areas with shopping malls. But these are too sterile and corporate — but they are also climate controlled and safer and caused the downfall of many downtowns. Then big box stores, more accessible from the burbs, kept at that tack.
The robotaxi will allow people to get more of what they sought from the “livable city” while still in sprawl. It will also let them get more of what they sought from the suburbs, in terms of safety and options for their children. They may still build pleasant pedestrian malls in which one can walk and wander among interesting things, but people who live 5 miles away will be able to get to them in under 10 minutes. They will be delivered right into the pedestrian zone, not to a sprawling parking lot. They won’t have to worry about parking, and what they buy could be sent to their home by delivery robot — no need to even carry it while walking among shops. They will seek to enjoy the livable space from 5 miles away the same way that people today who live 4 blocks away enjoy those spaces.
But there’s also no question that there will continue to be private malls trying to meet this need. Indeed the private malls will probably offer free or validated robotaxi service to the mall, along with delivery, if robotaxi service is as cheap as I predict it can be. Will the public spaces, with their greater variety and character be able to compete? They will also have weather and homeless people and other aspects of street life that private malls try to push away.
The arrival of the robocar baby-sitter, which I plan to write about more, will also change urban family life. Stick the kid in the taxi and send him to the other parent, or a paid sitter service, all while some adult watches on the video and redirects the vehicle to one of a network of trusted adults if some contingency arises. Talk about sending a kid to a time-out!
Here’s a nice graphic showing traffic deaths around the world. Of course, all these numbers are going to drop over the next 10 years thanks to various collision avoidance and accident survival technologies in cars, and eventually, we hope, robocars.
I was recently approached by a programmer named Keith Curtis, formerly at Microsoft and now a FOSS devotee. He wants to develop a driving simulator for testing robocar systems. I think this is a very worthwhile idea — sort of a “Second Life” for robots. We have a head start — the world of racecar video games has already done a lot of the legwork to simulate driving, and there are two open source car racing systems.
A good simulator would bring some tremendous benefits to robocar development.
Anybody, even a solo hacker in their basement, could develop and test robocar software on the cheap, and with no cost and risk from crashes. Small teams, perhaps working in car-less garages, could contribute greatly to the field.
Testing could go faster, and include regular exposure to extreme situations not often found in the real world, like crazy drivers, strange hazards, map errors, sensor failures and more.
Simulator testing allows the creation of new sensors which are plausible, but too expensive or too far in the future to work with in the real world. It allows teams to say, “What if we had 1cm accurate GPS? What if we had 180 line LIDAR to 100m?” and see if they can build robocar controls to use it.
Robocar contests could be held in simulation, on the cheap and with no physical risk. The winners could then get funding to build real-world vehicles to race for a bigger prize.
The simulator APIs for car controls and sensors can become prototype APIs for real-world interfaces, allowing easy testing and switching.
Of course, robocar simulation is nothing new. Several teams in the DARPA challenges built simulators to try out ideas. These remained proprietary efforts. Road simulation is also frequently used for traffic simulators. An open simulator would be shared, and the community (and not just the robocar development community) could contribute terrain, streets, sensors and simulators for elements such as pedestrians, human driven cars, blowing trash and new sensors, to name just a few.
Our wonderful new fast GPUs will be able to generate camera views anywhere in the 3D world for those working on machine vision. Simulating LIDAR, radar, ultrasound, odometry, accelerometers etc. is not yet done in car racing games but these things should not be hard to add. Indeed, any company selling a sensor would be well advised to build a simulated version of it.
And people hacking at home love to make 3-D maps of terrain. Existing real terrain models could be imported, or new ones made by driving around with LIDAR on real streets.
To explore this more I have written a new article on how to build a robocar driving simulator where I also point to an up and coming open source simulator called “Rigs of Rods” which actually simulates the vehicles at the physics level, treating them as a network of many connected parts.
The robocar world needs somebody ready to fun the kick-starting of such a simulator, and possibly some contests within it.
Tomorrow, I will be speaking on pre-Robocar technology at BIL an unconference that parallels the famous and expensive TED conference. This is in Long Beach, CA. Unconferences are fun, cheap and often as good as expensive conferences. I will also be attending a reception at TED tonight for Singularity University, which I lecture at, so I may see you if you’re at TED as well.
Last night’s EFF bash was a great success. Thanks to Adam Savage and all the others who made it go so well.
This weekend is the Foresight Institute conference on molecular nanotechnology and AI. I am on the board of Foresight Institute and will be speaking on the latest developments in robocars at the conference, along with a raft of great speakers. If you are interested in futurist issues around AI, nanotech and other accelerating technologies, this is the longest running conference in the field and the place to be. The whole conference is just $175 and you can register for it here.
I will also be doing my general Robocar talk on Wednesday, February 24th at the “Homebrew Robotics Club” of Silicon Valley. This is a great group of people who hack robotics as a hobby, and it meets at the CMU building at NASA Ames Research Center. This event is free and open to the public.
Finally, in 2 weeks I will be attending the DLD 10 conference in Munich Germany, and may try to make my way into some parties at the World Economic Forum in Davos while doing a little Alpine road trip after DLD.
You may have seen it already but it’s amusing to watch this encoding of a 1958 Disney show on the highway of the future:
This highway features a mixture of human driven cars, robocars and PRT style robocars on private guideways. Much of it is typical of ancient predictions of the future, with an expectation of remarkably cheap and strong materials and portable atomic power, but some of it is on the mark (like urban sprawl.) The roles of mom and dad don’t change in 50 years. It is always humbling to go back into the past and see how futurists have got it wrong, and wonder where you are going wrong in your own predictions. (I’ve learned you should never predict dates for things because while you might have a sense about how long it would take to develop the technology, you can’t as easily predict how markets and governments will react.)
However, I will be prognosticating again next week, giving my Robocar talk in the Google “Tech Talk” series at Google HQ in Mountain View on Dec 14 at 11am. While not generally open to the public, I can bring in guests if they all come in at around 10:30am. Contact me if this is of interest. They will put the talk up on Google Video when done. Of course, if you are a Googler, I hope to see you there.
It’s amusing to note that many of the vision seen in the movie “Minority Report” are found in this Disney video from much earlier.
I struck a nerve several years ago when I blogged about the horrible beep-beep noise made by heavy equipment when it backs up. Eventually a British company came up with a solution: a pulsed burst of white noise which is very evident when you are near the backing up vehicle but which disperses quickly so it doesn’t travel and annoy people a mile away as the beeps do.
Now I am seeing more and more suggestions that electric cars, which run quite silently when slow, make some noise for safety. This is fine, but there are also suggestions that there will be music and vanity noises, like ringtones or “cartones.” I can certainly see why this would appeal to people. (Already many think that their car is the place to play mind-numbing bass to announce musical taste to all others on the street.) There are even proposed laws.
While the cartones would be quieter than the backup beep or the heavy bass, I really fear that people will overdo what they think is the purpose — being attention grabbing. They will want to distract, and that will create a cacophony on the roads. It’s hard to make sounds that are meant to be attention grabbing (or vanity oriented) not travel beyond the range that you need them for safety.
I don’t want to imagine what it might be like living as I do with a 3-way stop outside my window, with each car singing a different tune or strange noise every time it slows down and starts up again. Who will want to live near intersections or parking lots?
I have a few proposals:
Like the beep-beep solution, use white noise that just doesn’t travel very far, but is easily noticed when close.
Use natural sounds, like waves crashing, birds chirping, wind blowing. We are tuned to hear those sounds in an otherwise silent environment, but our brains also can easily ignore them in background form.
Do indeed tune the volume based on ambient noise. This is suggested in the O’Reilly article linked above. They propose it to be loud enough. It should also be quiet enough.
Don’t do it at a speed where the tires and wind and electric motors are making enough noise already.
As robocar sensors become more common, such as LIDAR and radar, only make the noise when there are people who might come in contact with the vehicle. Otherwise, be silent.
Since robocars will not hit people in any normal operation, even people who don’t know they are there, such vehicles need not make any noise. HOwever, if they see a human or anything else on a collision course, let them make a more loud and useful noise that really gets attention, like a burst of white or pink noise, or even a horn if they ignore that. Start quiet, get louder if it is not reacted to in a human reaction time.
Let’s not give up on this opportunity to return peace to our public spaces as electric cars and robocars become popular.
A proposal is being floated in Europe for computerized convoys or road trains within the next decade. This is a proposal for a system where cars can hand over control to a lead car and follow in a train or convoy, without physical connection.
This idea comes up a lot as an early robocar technology. It is particularly common because it’s much easier to do — a human driver still is in charge, and the robotic control is limited to a very limited and simple environment. It’s safe to say that we could make this work very quickly if we wanted to. There is no navigation or vision required, no recognition of obstacles, no choice of speeds or turns. Cars that come together in a convoy can draft to get a serious boost in fuel efficiency, and of course the un-drivers can now relax and read or work on the trip.
As a robocar booster, people are surprised when I say I am not too thrilled about this idea, at least as an early technology. Rather I think it’s a great idea for later. In spite of the enthusiasm with which I write, the robocar problem is not a simple one. This much simpler problem is tempting but has some snags.
First of all, if you have a bug in a standalone robocar system, it may cause an accident, and that may injure or kill the occupants of the robocar, and perhaps one or two other cars. Death is less likely at urban speeds of course. A problem with a computerized convoy could have terrible results, involving scores of cars. Since most people want this for the highway, the problem would also occur at lethal speed. Convoys are just not the first place we want to test our systems and have our first accidents.
Secondly, forming convoys requires a critical mass of suitably equipped cars. Of course, you don’t need a dozen full robocars to make a train, all you would need is cars with drive-by-wire and some much simpler control circuitry. But even so, the incentive to get a car with this feature has to get over a critical mass hump if it’s going to be worthwhile. It’s not quite as bad as fully ad-hoc trains, since you can have scheduled trains, lead by a bus or truck driver, and cars can see such a lead vehicle and get in behind it. But at first, the odds of many cars all finding one another at the same time is low. If the train is going faster than regular traffic in a carpool lane, as we hope it would, it will not be easy to join a train that moves past you on the highway. If it moves slower than traffic, it is easy to slow down and join it, but then it has to move slower, with all the attendant problems.
Computerized convoys have advantages and disadvantages over physical ones. Physical ones probably can only be formed while stopped, and probably only unformed that way too. One could see the last car in a physical convoy undocking while moving, so with correct ordering it might work out, but it’s a far cry from a virtual convoy which allows anybody to join and leave at any time.
Physical convoys however can transmit power. This is useful if you expect people to be driving short-range electric cars. They would take their short range car and join a convoy, and be powered by the lead locomotive while operating, and even be recharging a bit. After dispersion, the vehicles would only need to go a short destination to their target and back to the evening train.
Physical coupling makes it harder for one car to leave the train due to a failure. On the other hand it means that if the lead car wants to change lanes, all cars must do so. If the lead car leaves the road, they all do. Jack-knifing is a real worry, which is one reason that today even cargo road trains are limited to 2 trailers in urban areas, and 3 trailers in rural areas, if they are allowed at all.
Physical coupling requires specially modified vehicles. This is even more the case if the locomotive will actually be towing the vehicles physically rather than providing them with electricity for their motors and batteries. Either of these is a major modification, while virtual coupling only requires a drive-by-wire car and a small matter of programming.
Even full robocars probably should not form convoys right away. We should wait until our confidence is even higher, in spite of the fuel savings. If one car goes bad, or its occupants try to take over and move to manual driving, the consequences could be nasty in any convoy. And of course, the first robocars on the road will never get to join convoys as they will not meet the others. That’s why you need to solve the solo navigation problem first, and then you get enough on the road to work on the cooperation problems.
First: I will be speaking on robocars tomorrow, Tuesday Nov 9, at 6:30 pm for the meeting of the Jewish High Tech Community in Silicon Valley. The talk is at 6:30pm at the conference center of Fenwick and West at Castro & California in Mountain View. The public is welcome to attend, there is a $10 fee for non-members. This will be similar to my talk at Stanford 2 weeks ago, and a bit more extensive than the one in New York early in October, which Forbes said was the audience favorite at the event.
Volvo has built a brand around safe cars, and last year committed that nobody would die in a newer Volvo by 2020. They plan to do much of this with better passenger safety systems, akin to the work they have done on airbags and crumple zones. However, they also intend to use a lot of computerized technologies to make it happen. Other teams are pushing to expand the goal inside Volvo to also stop people from being killed by Volvos. To that end, next year’s Volvo S60 will come with a “Pedestrian Avoidance System” which uses a camera and machine vision to identify pedestrians and calculate if the vehicle is about to hit one. If it sees a potential pedestrian collision it will beep and alert the driver. If the driver does nothing, the car will brake.
Here is a video of the S60 in action:
It’s impressive, though pure machine vision suffers problems as lighting changes, which is one reason most work recently has been on LIDAR. It’s also interesting to see if they will be able to avoid making it too conservative. If the warning goes off all the time, even for a pedestrian who will (to the human eye) clearly slide by the side of the car at places like a crosswalk, drivers may learn to ignore the alarm, or get very annoyed and shut the whole system off it it brakes for them when they know an impact is not imminent. I’m hoping to learn more about Volvo’s efforts in the future. No other company has put as much effort into building a brand around safety, so we can expect Volvo, which has slipped in this status of late, to work very hard to maintain it and adapt robocar technologies to safer human driving and fully autonomous driving
Dense triple parking
I have written of a simple algorithm to allow dense Valet style parking of robocars, such as triple parking on the roadsides. In this algorithm, one gap is left in the outer lanes, and the Robocars are able to move together, as an entire row segment, to “move the gap” as quickly as a single car can move. That way, if a car needs to get out from an inner lane, it can signal, and if the gap is currently ahead of it, for example, all the cars from the one next to it to the gap can move forward one space (at the same time) to put the gap next to the vehicle that needs to leave. This can happen in all the other rows and is easy, quiet and efficient for electric cars. It does not even need radio communication, as robocars will sense a car moving behind them or ahead of them, and immediately move in reaction. This request will move up the chain of cars to the gap. Of course, if one car does not move, the car behind it will only move a very short distance before refusing to go further, which would stop the whole effort (or in the case of an error, cause a very slow impact if the car behind keeps coming) and signal a need for human attention.
It seems like this should be possible even without many gaps, as long as there is enough spare space to allow a vehicle to wiggle out of its space. If there is just one gap, and a bit of wiggle room in the other rows, any car can still get out, just a bit more slowly. This is probably better done with a protocol for communication to assure it works quickly.
In this case, a gap on the outside lane (where there must be at least one) can be temporarily moved to the inside, and then back out. Consider 3 lanes of cars, with a gap in the outer late (lane #3) and a car in lane #1 (the curb lane) wanting out. First the lane #3 cars would adjust to move the gap to the right place, a bit forward of our target car. Next, a car from lane #2 would move into this gap, leaving a gap in lane #2 into which our target car can move. This leaves a gap in lane #3 which can be filled by a car from lane #2 which is willing to move in, ideally right next to our target car. Likewise a car from lane #3 can now move into that gap, and the resulting gap in the outer lane #3 can be moved to allow exit by our target car.
This requires a great deal more car moving, though again with electric cars this may not be too expensive. If the cars can turn all their wheels, they can move horizontally as some concept cars can already do. Even without that, a robotic car can wiggle out without much room, and of course the gap would not be placed exactly in place with the target car, but probably slightly forward to allow transfer with fewer wiggles. The result is a whole valet lot with just one blank space needed to get any car reasonably quickly. Of course, this would only be done when the lot needed to be totally full. For any partially full lot, gaps would be left to minimize the car moves needed to get any car out. However, if space is at a premium — so much so as to justify the extra moving — it can be done.
While giving a talk on robocars to a Stanford class on automative innovation on Wednesday, I outlined the growing problem of software recalls and how they might effect cars. If a company discovers a safety problem in a car’s software, it may be advised by its lawyers to shut down or cripple the cars by remote command until a fix is available. Sebastian Thrun, who had invited me to address this class, felt this could be dealt with through the ability to remotely patch the software.
This brings up an issue I have written about before — the giant dangers of automatic software updates. Automatic software updates are a huge security hole in today’s computer systems. On typical home computers, there are now many packages that do automatic updates. Due to the lack of security in these OSs, a variety of companies have been “given the keys” to full administrative access on the millions of computers which run their auto-updater. Companies which go to all sorts of lengths to secure their computers and networks are routinely granting all these software companies top level access (ie. the ability to run arbitrary code on demand) without thinking about it. Most of these software companies are good and would never abuse this, but this doesn’t mean that they don’t have employees who can’t be bribed or suborned, or security holes in their own networks which would let an attacker in to make a malicious update which is automatically sent out.
I once asked the man who ran the server room where the servers for Pointcast (the first big auto-updating application) were housed, how many fingers somebody would need to break to get into his server room. “They would not have to break any. Any physical threat and they would probably get in,” I heard. This is not unusual, and often there are ways in needing far less than this.
So now let’s consider software systems which control our safety. We are trusting our safety to computers more and more these days. Every elevator or airplane has a computer which could kill us if maliciously programmed. More and more cars have them, and more will over time, long before we ride in robocars. All around the world are electric devices with computer controls which could, if programmed maliciously, probably overload and start many fires, too. Of course, voting machines with malicious programs could even elect the wrong candidates and start baseless wars. (Not that I’m saying this has happened, just that it could.)
However these systems do not have automatic update. The temptation for automatic update will become strong over time, both because it is cheap and it allows the ability to fix safety problems, and we like that for critical systems. While the internal software systems of a robocar would not be connected to the internet in a traditional way, they might be programmed to, every so often, request and accept certified updates to their firmware from the components of the car’s computer systems which are connected to the net.
Imagine a big car company with 20 million robocars on the road, and an automatic software update facility. This would allow a malicious person, if they could suborn that automatic update ability, to load in nasty software which could kill tens of millions. Not just the people riding in the robocars would be affected, because the malicious software could command idle cars to start moving and hit other cars or run down pedestrians. It would be a catastrophe of grand proportions, greater than a major epidemic or multiple nuclear bombs. That’s no small statement.
There are steps that can be taken to limit this. Software updates should be digitally signed, and they should be signed by multiple independent parties. This stops any one of the official parties from being suborned (either by being a mole, or being tortured, or having a child kidnapped, etc.) to send out an update. But it doesn’t stop the fact that the 5 executives who have to sign an update will still be trusting the programming team to have delivered them a safe update. Assuring that requires a major code review of every new update, by a team that carefully examines all source changes and compiles the source themselves. Right now this just isn’t common practice.
However, it gets worse than this. An attacker can also suborn the development tools, such as the C compilers and linkers which build the final binaries. The source might be clean, but few companies keep perfect security on all their tools. Doing so requires that all the tool vendors have a similar attention to security in all their releases. And on all the tools they use.
One has to ask if this is even possible. Can such a level of security be maintained on all the components, enough to stop a terrorist programmer or a foreign government from inserting a trojan into a tool used by a compiler vendor who then sends certified compilers to the developers of safety-critical software such as robocars? Can every machine on every network at every tool vendor be kept safe from this?
We will try but the answer is probably not. As such, one result may be that automatic updates are a bad idea. If updates spread more slowly, with the individual participation of each machine owner, it gives more time to spot malicious code. It doesn’t mean that malicious code can’t be spread, as individual owners who install updates certainly won’t be checking everything they approve. But it can stop the instantaneous spread, and give a chance to find logic bombs set to go off later.
Normally we don’t want to go overboard worrying about “movie plot” threats like these. But when a single person can kill tens of millions because of a software administration practice, it starts to be worthy of notice.
Saturday saw the dedication of a new autonomous vehicle research center at Stanford, sponsored by Volkswagen. VW provided the hardware for Stanley and Junior, which came 1st and 2nd in the 2nd and 3rd Darpa Grand Challenges, and Junior was on display at the event, driving through the parking lot and along the Stanford streets, then parking itself to a cheering crowd.
Junior continues to be a testing platform with its nice array of sensors and computers, though the driving it did on Saturday was largely done with the Velodyne LIDAR that spins on top of it, and an internal map of the geometry of the streets at Stanford.
New and interesting was a demonstration of the “Valet Parking” mode of a new test vehicle, for now just called Junior 3. What’s interesting about J3 is that it is almost entirely stock. All that is added are two lower-cost LIDAR sensors on the rear fenders. It also has a camera at the rear-view mirror (which is stock in cars with night-assist mode) and a few radar sensors used in the fixed-distance cruise control system. J3 is otherwise a Passat. Well, the trunk is filled with computers, but there is no reason what it does could not be done with a hidden embedded computer.
What it does is valet park itself. This is an earlier than expected implementation of one of the steps I outlined in the roadmap to Robocars as robo-valet parking. J3 relies on the fact the “valet” lot is empty of everything but cars and pillars. Its sensors are not good enough to deal well with random civilians, so this technology would only work in an enclosed lot where only employees enter the lot if needed. To use it, the driver brings the car to an entrance marked by 4 spots on the ground the car can see. Then the driver leaves and the car takes over. In this case, it has a map of the garage in its computer, but it could also download that on arrival in a parking lot. Using the map, and just the odometer, it is able to cruise the lanes of the parking lot, looking for an empty spot, which it sees using the radar. (Big metal cars of course show clearly on the radar.) It then drives into the spot.
Some time ago I proposed the “School of Fish Test” as a sort of turing test for robocars. In addition to being a test for the cars, it is also intended to be a way to demonstrate to the public when the vehicles have reached a certain level of safety. (In the test, a swarm of robocars moves ona track, and a skeptic in a sportscar is unable to hit one no matter what they do, like a diver trying to touch fish when swimming through a school.)
I was interested to read this month that Nissan has built test cars based on fish-derived algorithms as part of a series of experiments based on observing how animals swarm. (I presume this is coincidental, and the Nissan team did not know of my proposed test.)
The Nissan work (building on earlier work on bees) is based upon a swarm of robots which cooperate. The biggest test involves combining cooperating robots, non-cooperating robots and (mostly non-cooperating) human drivers, cyclists and pedestrians. Since the first robocars on the road will be alone, it is necessary to develop fully safe systems that do not depend on any cooperation with other cars. It can of course be useful to communicate with other cars, determine how much you trust them, and then cooperate with them, but this is something that can only be exploited later rather than sooner. In particular, while many people propose to me that building convoys of cars which draft one another is a good initial application of robotics (and indeed you can already get cars with cruise control that follows at a fixed distance) the problem is not just one of critical mass. A safety failure among cooperating cars runs the risk of causing a multi-car collision, with possible multiple injured parties, and this is a risk that should not be taken in early deployments of the technology.
My talk at the Singularity Summit on robocars was quite well received. Many were glad to see a talk on more near-future modest AI after a number of talks on full human level AI, while others wanted only the latter. A few questions raised some interesting issues:
One person asked about the insurance and car repair industries. I got a big laugh by saying, “fuck ‘em.” While I am not actually that mean spirited about it, and I understand why some would react negatively to trends which will obsolete their industries, we can’t really be that backwards-looking.
Another wondered if, after children discover that they nice cars will never hit them, they then travel to less safe roads without having learned proper safety instincts. This is a valid point, though I have already worried about what to do about the disruption to passengers who have to swerve around kids who play in the streets when it is not so clearly dangerous. Certain types of jaywalking that interfere with traffic will need to be discouraged or punished, though safe jaywalking, when no car is near, should be allowed and even encouraged.
One woman asked if we might become disassociated with our environments if we spend our time in cars reading or chatting, never looking out. This is already true in a taxicab city like New York, though only limos offer face-to-face chat. I think the ability to read or work instead of focus on the road is mostly a feature and not a bug, but she does have a point. Still, we get even more divorced from the environment on things like subways.
As expected, the New York audience, unlike other U.S. audiences, saw no problem with giving up driving. Everywhere else I go, people swear that Americans love their cars and love driving and will never give it up. While some do feel that way, it’s obviously not a permanent condition.
Some other (non-transportation) observations from Singularity Summit are forthcoming.
BTW, I will be giving a Robocar talk next Wednesday, Oct 28 at Stanford University for the ME302 - Future of the Automobile class. (This is open to the general Stanford community, affiliates of their CARS institute, and a small number of the public. You can email firstname.lastname@example.org if you would like to go.)
Today’s systems are fairly simple of course, and will learn a lot from this. This matches my prediction for how a robocar test suite will be developed, by gathering millions and later billions of miles of sample data including all accidents and anomalous events, over time with better and better sensors. Today’s sensors are very simple of course but this will change over time.
Initial reaction to these systems (which will have early flaws) may colour user opinion of them. For example, some adaptive cruise controls reportedly are too eager to decide there is a stopped car and will suddenly stop a vehicle. One of the challenges of automatic vehicle design will be finding ways to keep it safe without it being too conservative because real drivers are not very conservative. (They are also not very safe, but this defines the standards people expect.)
Tonight, at 111 Minna Gallery in San Francisco, we at EFF will be hosting a reading by Randall Monroe, creator of the popular nerd comic “xkcd.” There is a regular ticket ($30) and a VIP reception ticket ($100) and just a few are still available. Payments are contributions to the EFF.
In two weeks, on Oct 3-4, I will be speaking on the future of robot cars at the Singularity Summit in New York City. Lots of other good speakers on the podium too.