Brad Templeton Home
Roadmap to Robocars
In Part 1 - The case for Robot Cars I outlined just how important it is that we develop autonomous self-driving robot cars as quickly as we possibly can. Teams of engineers and scientists are hard at work on this development end of this project, and the outlook is optimistic.
But there are other "roadblocks" that stand in the way. Social roadblocks. Particularly in the USA, the most car-loving nation on the planet. Indeed, in spite of the USA being a major source of the innovation, robocars (as I am calling them) may first be deployed in other nations like India, Japan, China or Germany.
Here I present a list of both existing and hypothetical technologies which might take us, one step at a time, to the world of robocars on city streets.
What's Already Here
You may not realize how many computer-assisted driving technologies are already in the market or announced.
Near Term Technologies
These available-now features are all examples of incremental technologies that will be marketed as making a car safer and more accident resistant. Most robocar technologies will see their first non-military on-road deployment in this fashion. Luxury car makers will be able to add plenty to the price to make a car which is "crash resistant." Over time, the technologies will make it into cheaper and cheaper cars.
This is doable because according to an NHTSA study, 80% of accidents are caused by driver inattention. Systems that can see hazards that drivers aren't paying attention to, and warn them, may prevent a great number of accidents.
Some years later, we should see an almost "crash-proof" car. This will be a car that is very difficult to crash, even if you try. It need not interfere much with your driving, in that what it has to determine is whether the car is approaching a situation where the computer, even with its superior senses and reaction times, can't get out of. It will resist the driver who tries to approach those boundaries, though it probably won't stop a driver who does a forceful override. (ie. the wheel will get very hard to turn if moving into a danger zone, but if you turn it hard enough, it will obey you. For the remainder of your life, perhaps.)
The main question will be how much to tune that buffer zone. The closer you can get to the danger line without triggering the system, the less the system will be visible to the driver. Indeed, ideally a driver who never would have gotten into an accident will never even notice the system, because there will be very few situations the human could get out of that the computer could not.
Such systems would probably have to ignore oncoming traffic that stays in its own lanes. While oncoming vehicles present an obvious danger of creating a situation (head-on collision) that nobody (human or robot) can get out of, in our driving habits even the most defensive driver just accepts that. Having one-way streets and more divided roads solves this problem.
This car eventually should qualify as a full robocar; when heading into an accident, the driver could simply release the controls, and the computer will plot the courses of all other vehicles and obstacles and do the best possible job to bring everything to a safe condition. If the human gives guidance, the computer would follow it unless the human is guiding towards a situation that can't be recovered from.
The main thing stopping that vehicle from having full robocar status would be its general ignorance of the other basics of driving and navigation. It will be concerned entirely with short term factors and escape from catastrophe.
It's easy to see concerned parents paying large premiums for such cars for their children, and insurance companies offering serious discounts to owners of these vehicles.
At some point we'll have cars that can drive themselves, all or most of the time, but it won't be legal to let them. The technology will be still permitted as anti-accident technology, since if the driver is knocked out or falls asleep and lets go of the wheel, people will want a car that's able to handle the situation. But as confidence grows, people will (perhaps illegally) let go of the wheel more often.
History shows, however, that car safety technologies tend to often just cause drivers to feel safer and take greater risks, sometimes restoring, or even increasing the accident rate. Predictions for these technologies will be difficult to make.
I should point out that this technology does counter one of the main statements I've made here. I've written that each year we delay robocar technology, human drivers will kill 45,000 people in the USA and a million around the world. Crash-avoiding cars can change that, in that they should drop the death toll as they get deployed, or at least render fatal events non-fatal.
Some may wish to stop development at crash-avoiding cars, but almost all the technologies in that marketplace will also enable the robocar.
Volvo has promised to deliver an injury-proof car by 2020 that uses both crash avoidance, and better passenger protection systems to deal with crashes it can't avoid. They think locusts might offer a clue.
Human drivers are demanding and buying navigation systems which will be the prototypes for robocar navigation. Features coming include highly accurate assisted GPS, dynamic traffic data gathered from other cars and corrections based on observing user behaviour and user reports. Eventually such systems will start understanding even lanes and parking lots, as well as traffic light timings, congestion charging and more.
There is an entire field known as "Intelligent Transportation Systems" devoted to concepts of networking cars for better navigation and traffic management. Much of their work will also facilitate robocars, as robots are frankly a lot better at constantly integrating such data into their decision making.
Simulated Robocar Environment
I have joined others in proposing the development of a networked open simulator platform for robocars which would allow small teams, at minimal cost to develop and test robocar software, and improve on other's open source efforts. This environment could also be used for contests, and the winners could then get funding to try their software in real vehicles.
Thanks to the DARPA grand challenges, and the congressional mandate that 1/3 of all miltary vehicles be autonomous by 2015, the first autonomous cars will be military delivery robots, or "deliverbots." These will be deployed outside the USA, typically in danger zones, to move cargo without risk to human drivers from enemy attack. Some cargo will be destroyed or possibly stolen, but overall the cost will be vastly less.
Some cargo -- like weapons -- is so valuable it must still be guarded. However, those convoys will consist of robots transporting the weapons, and heavily armoured, bomb-resistant manned vehicles escorting them.
The military technology will produce many spin-offs for civilian use. In addition, the civilian deliverbot is a world-changing concept on its own.
Telemetry & Networking
Robocars ideally will operate completely independent of any other systems. They should not depend on any central control or broadcasts, nor should they depend on talking to other robocars. However, that doesn't mean they can't perform better by using such information, and they will.
Even before robocars are on the road, HDVs (human driven vehicles) can benefit from this sort of information. For example, it is useful to receive broadcasts to know when traffic lights will change, so that you can time your driving to not hit red lights. In fact, studies show that if a car is driven this way, and thus does not stop and start, it results in a 30% improvement in city mileage -- about the same benefit a hybrid car gets from using electricity for starts and stops. (As noted above, Audi Travolution already does this.)
This is already underway. A protocol known as 802.11p has been defined for WAVE (Wireless Access in the Vehicle Environment.) I've proposed an option for the gas pedal where it redirects your attempts to go pointlessly faster just so you can end up stopped at a light -- unless you really insist.
Information on live traffic patterns is also useful for deciding the best route for a trip. It is not necessary for any central control to tell cars where to go -- they will naturally want to take the less congested, faster roads. Cars may announce to the network that they expect to be at certain places so that other cars can decide to avoid those places if the counts will cause too much congestion.
Telemetry from other cars is particularly useful to robocars. If a robocar can know that another car is one of its kin -- and thus has a computer's fast reaction times -- it becomes possible for the cars to swerve en masse around obstacles, pedestrians or rogue HDVs. Of course, a robocar must not depend entirely on such information, but if you decide you can trust another vehicle, cars can pack more closely on the road to increase road capacity.
Other useful telemetry can include reports on potholes and traction reports, inch by inch, on wet and icy roads. This can also be used by an HDV to improve vehicle safety and performance.
If you like, you can jump ahead to more thoughts on the "motornet."
That self-parking Lexus is already for sale. This does parallel parking in a small area. Imagine going a step further, and allowing a car to park itself in a special section of a parking facility meant for such operations. A car equipped with "Robovalet" might be able to obey commands from the parking area, following markers or (in early versions) guide wires set in the road. The robovalet area would not encounter any "civilians" (ie. non-employees) and would be on private property, allowing it to operate long before robocars are allowed on the road. It would be more like the robots that already roam factory floors and hospitals.
Robovalet cars might park themselves densely, and move out of the way on command. They could self park in garages with limited height. It's a vastly simpler problem than full street driving.
Robovalet cars could also be programmed to form a convoy where a large series of cars would follow one another. This would allow a single human valet to take a large group of cars from the drop-off area to a more remote long-term valet parking lot at an airport -- only one human would be required to move a dozen cars. The human valet could lead back another train of cars for those passengers who are slated to arrive soon. The trip does not have to be done at high speed, so it can be quite safe even on ordinary roads. We only have to trust the car to follow the leader -- something we can already attain today.
People would happily pay more for a car that they can just drive to the Robovalet entrance at the airport, office or mall, and then get out and leave to itself. Arriving back, they would signal on their phone and find their car ready and waiting to drive away.
It is possible that the best way to do this would be to not insert the full Robovalet logic into each car. Instead, each drive-by-wire car would come with a standard interface which would allow a "robovalet module" to be mounted on the car. This module, equipped with the latest technology, and access to any sensors in the car, would belong to the parking lot and do the real work. A standard (perhaps wireless) would be created to allow the owner to grant control of their car to the valet module.
One could even imagine a valet module which works on older cars with less drive by wire. Almost all cars have drive by wire throttle (cruise control) and brakes now, which could be enough if the valet module has a small "tug" which provides steering guidance to the car, while commanding the car to provide the real power and braking. If brake control can be done independently on the wheels, you can even perform limited steering with just brakes.
In the extreme case, one could build a full tug module that you drive 2 wheels of the car onto (like a tow truck) for cars that are not themselves equipped to handle the Robovalet protocol.
A more advanced version of the robovalet -- the self-delivering car -- might come next to provide many of the advantages of the robocar.
We could also see cars more designed for compact robo-valet parking, like the MIT proposed City Car.
This step is about to move into the "implemented" column. Stanford's robotic car team has modified a Volkswagon known as Junior 3 to be able to find an empty parking spot in a parking lot and park itself in it, if given a map of the lot.
Car-style Personal Rapid Transit
While you'll find in other articles that I predict that PRT may be obsolete before it gets wide use, the earliest successful ventures in PRT are actually along the Robocar direction. The parking shuttle at Heathrow's Terminal Five is a rubber-tire based PRT called ULTra. ULTra still runs on dedicated right-of-way with small curbs that it follows, as it can't detect and avoid pedestrians or go an anything but its appointed route. This system entered test service for the airport employees in 2010 and is scheduled to enter full public revenue service very soon.
Abu Dahbi is building a brand-new city near its airport called Masdar. Masdar is intended to be a car-free city, and as such they are building the pedestrian part of the city one floor up. On the ground -- the apparent basement of the buildings -- are tracks for robotic cars for personal transportation and cargo shipping. These cars follow magnetic tracks buried in the ground rather than curbs, and do not encounter pedestrians or human driven cars, though they do have a laser detector which stops them if a pedestrian or other obstacle is in the way. This system (and the whole city plan) was scaled back with the economic downtown, but is now in operation with 2 stops for passengers and 3 for cargo.
Like all PRT systems, these systems require dedicated right-of-way. But they will act as useful demonstrations and get the public more aware of robot cars. By running on tires they have the ability to work on a more standard infrastructure as they improve, unlike custom rail-based PRT which is tied to its track.
Airports are one of the areas where PRTs show the greatest chance of success. Heathrow's demonstration may encourage other airports to try even more advanced technologies, such as robot cars that follow planned tracks but which have LIDAR to assure no harm to predestrians. Along a planned track, one could even have redundant sensors for pedestrians and other vehicles, some in the track and some in the cars.
A possible alternative to the Robovalet is the Televalet, which would include a set of cameras to show a 360 degree view from the car. This video would be routed to a remote driver sitting at a console, driving the car in a parking lot or on parking routes -- almost like a video game.
This is really not too far fetched for a drive-by-wire car as long as speeds are modest and the car provides safety sensors that stop it if it appears to be close to hitting something. (Though such impacts would be fender benders, and not put humans at risk.)
The Televalet, like the Robovalet, allows more efficient parking, especially at large offices and airports.
The military also has a history of remotely controlled robot vehicle technology. Flying vehicles (UAVs) are the rage right now but long range ground vehicles will also be desired. (The main problem they face is a reliable control signal in an urban environment without a solid RF data network. They might solve that with repeaters in the air and limited autonomy during loss of signal.) The military, of course, would also like to arm such vehicles.
If the network connection fails, the vehicle would need to be able to maintain a steady state, and then pull over and stop quickly to maintain safety. Today's network technology is not quite reliable enough for this but other forces are working to bring it there.
Medium Term Technologies
The next big step is an autopilot. Like an aircraft autopilot, it would drive the vehicle in simple situations, but a human driver would always be watching and ready to take over. The autopilot would be capable of keeping in highway lanes, maintaining proper spacing from other cars. It would be able to sound loud alarms if anything unusual should happen.
The simple autopilot is a robust version of what we see now for highway driving. In fact, vendors selling lane departure warning system say the main reason they don't actually stop the car from leaving the lane is liability. Along with the surprise to the user who forgot to signal, I suppose.
The autopilot would not change lanes, or turn on its own, though it might handle highway to highway interchanges. It might also be able to handle urban one-way streets though the presence of pedestrians raises the bar a fair bit.
(This is not so far away, if you read about the 1995 VaMP autopilot experiment.}
Human assist autopilot
Another variant of autopilot would be a system that needs occasional advice from a human, but does not need the human's full attention. (Systems that can drive with occasional correction from an attentive human were built over a decade ago.)
In this case, if the system sees a (rare) situation that it can't understand, it would seek human advice. It would slow or stop if need be until the advice arrived. Ideally the unknown situation would be detected well in advance. For example, if the car's database states a traffic light is present at an upcoming intersection, and it is unable to detect the light, this might be worth asking a human about.
(The human need not be in the car, the question could be resolved by sending video to a remote person.)
A departure from the above path which may come sooner than the full robocar is the mini delivery robot, or deliverbot. This would be shown to work first in military zones, but once trusted enough it could find its way to the cities.
Why would they let this on the street before a robocar?
On the other hand society may be quicker to trust the autopiloted car with a watchful human quasi-driver before they trust the deliverbot. However, if cars can't deliver themselves to people on demand, many of the great benefits are diminished.
I have an essay on other interesting consequences of the urban deliverbot.
Self-delivering car (Whistlecar)
Deliverbot level technology enables something even more interesting. A car that acts as a robocar only when it is vacant. In other words, we might find ourselves with a technology that is trusted to move vehicles at safe speeds on the roads, but not to carry passengers. This could be used to deliver vehicles to drivers on demand. Humans would still drive them when they get in.
I'm going to use the term "whistlecar" for these, as the idea is reminiscent of how the Lone Ranger could just whistle for his horse, Silver, and the horse would appear from him to ride. (Since these cars will come from even further away, perhaps Gandalf's steed Shadowfax is an even better metaphor.)
Whistlecars would deliver many of the advantages of robocars that I have outlined here, in particular the "car on demand" and "right vehicle for the trip." The whistlecars would be able to park and refuel themselves on their own, with all the advantages that implies.
They would not, however, allow the driver to work, read or relax. They would still need a "road feeling" suspension. They would probably not be as good at reducing congestion or drafting other vehicles. They could be two-wheeled.
And they could still get in accidents. However, they would presumably be packed with accident avoidance technologies and as such would probably get into far fewer. However, since they themselves would not save lives, any accidents they had while delivering themselves would be attacked as a complete net downside; that could impede their deployment.
Creating a whistlecar may be much easier to do, technically, than a robocar. While moving itself from place to place, it can move more cautiously and slowly. It can also brake and turn much harder if it detects a problem. It can even slow or stop and set up a video connection to a nearby televalet (remote operator) to solve problems only a person can solve. They might use only a very limited set of streets until close to their destination.
Whistlecars would take a little longer to get to you when summoned than a full robocar, but otherwise would still arrive quickly so long as there are enough of them out there. The whistlecar makes the Carshare concept considerably more appealing because the cars come to you. If you make an appointment, there would be no waiting.
You can read a more detailed examination of whistlecars.
We might see robot vehicles deployed first in rural environments, and in fact they already are in certain industrial cases. Mining, drilling and farming seem ripe for robotic equipment. These robots might also get early permission to travel on rural roads -- slowly and well marked and without people inside -- before we see full robocars.
Drafting improves gasoline mileage as much as 45%. That's because as you get faster, the power needed to overcome wind resistance, which goes up with the cube of speed, becomes the dominant energy sink of driving. Trains are efficient in part because there is one long train with a small wind profile.
Drafting other cars is not safe for drivers. If the vehicle in front brakes, a human can't react in time. An automated system could, once it is trustworthy. We do drive quite close on the highways, and most of the time without incident, but getting close enough for good drafting is much trickier.
However, we could go beyond that, and develop the concept of a convoy. In this train, the front driver is in control. All other cars follow the front car's command. While the cars are not physically connected, they could move as a train, almost doubling the efficiency.
Of course, one must really trust the lead driver. Such trains, which could form ad-hoc on the highways, might require the lead driver have a special licence, like a bus driver. (After all, we all trust ourselves to a bus driver's skills.) There could also be 2 or 3 drivers in the front who are actively at the controls, ready to override if the main driver does something erratic.
Cars equipped to "follow the leader" could see a road convoy go by, and query for permission to join it. They would go to the back (or even middle) of the train and yield control. A redundant net would be required, and in the event of loss of control all cars would move to a fail-safe mode, quickly increasing the gap or instructing their drivers to move to another lane. The cars would not need central control to maintain their distance from the other cars -- that technology is already available. The central control would just be for any (rare) lane changes, and to assure all vehicles stay in their lane.
Of course, we must be very confident of the safety of trains, since a failure here would be particularly catastrophic. Fear of this may delay adoption of convoys until much later.
Roll this truckin' convoy cross the USA
A convoy of robotrucks may come sooner than a convoy of passenger cars. Trucks are all driven by professional drivers who can be trained to lead convoys. In the Australian outback, they have "road trains" which are a truck with 3 trailers. These split apart at transfer stations on the outskirts of town. (A three-trailer road train tends to use 2/3 of the fuel and obviously 1/3 the staff of 3 one-trailer semi-trucks.)
In one simple mode, a convoy might form with 3 trucks with 3 drivers, each taking a shift in control of the convoy, letting the others relax or sleep. Other vehicles might join, to some limit. Because the trucks would draft one another, this would be much more fuel efficient than ordinary trucking, more competitive with rail (which currently beats trucks 4 to 1 on energy usage.) Of course the weight of the trucks demands top-flight safety on a system like this, and mechanisms to let cars pass these slow convoys easily.
There have in fact been experiments in Europe on such convoys.
The main problem with true interlinked road trains is safety. The vehicle is twice as long and 3 times as heavy. Robodriving might bring the safety level high enough to allow this on more roads, even with more trailers. Each trailer might also have a small engine to allow it to dock or even temporarily decouple in a danger situation.
As noted, it's quite possible that other countries may be the first to deploy robocars on real streets. Other nations, with their high fuel costs and poorer populations have been at the head of innovation in many automotive areas, especially small efficient cars.
While the USA loves cars more than any other country, and emits the most pollution from them, its legal and political system might stop it from leading the way. Japanese are well known for being more friendly to robot innovation. India and China have excellent high-tech workforces and much motivation. Traffic in India is vastly worse than in the USA. Shared cars make much more sense in poorer nations. Some nations, particularly China, can eliminate the political barriers overnight if the central government power brokers wish it.
Certainly if the cars are deployed, and proven safe in other nations, especially ones with chaotic streets, this should facilitate adoption in the USA. Indeed, nothing gets the USA going more than a feeling of being behind other nations in technology leadership. That is what inspired the first "Apollo project."
As confidence grows in the technology for accident avoiding cars and robocars, we'll see a few powerful forces work to lobby for them. Of course there should be new organizations convinced of their value, but consider these forces:
This is a powerful set of lobbyists. However, there will be opponents, such as oil companies, technophobes and unions such as the Teamsters, since truck and taxi drivers would be put out of work.
Longer Term Technologies
The School of Fish Test
We need our cars to be as smart as a fish.
On a test track, a swarm of robocars will be moving. A skeptic will be given the keys to a regular car -- sportscar or Hummer as they prefer -- and be told, "try to hit one." For the cars to pass this test, the driver will not be able to hit one.
Then, demonstrators will show how you can walk into the swarm and none of them will hit you, and in fact you won't be able to touch one, no matter how you try, leaping and diving, laying down in front of them and more.
How will the cars do this? They must identify every other vehicle, person and object on the road. They must predict where that person/vehicle could go if it acts normally, or does something erratic.
The skeptic's Hummer will be identified, and all possible paths it could take will be considered. The Robocar will just never put itself in a position where, if the human car did something erratic, the robocar, with its superior reaction times, could not evade it.
This is not that hard once the reaction times are very good. A robocar, able to notice the vehicle in front of it is braking (brake lights or not) could respond within a few milliseconds. Humans take about 750ms.
If the car can react within 20 milliseconds (1/50th of a second) and can brake as well as the vehicle in front, then at 100 mph, it can squeak by with just a 3 foot gap from the vehicle in front. We wouldn't want to time it that closely, but this should give you the idea of what we're talking about. Steering questions are similar.
What gets harder is heavy traffic. In heavy traffic you might know you can brake or veer away if the unknown vehicle makes a sudden turn, but if there is another vehicle blocking your way, you could be limited, unless you know that it also will veer. To help here, a trust protocol, where cars can query other cars about whether they would swerve in tandem, could still allow dense packing. Note that the vehicle would not need to use this protocol to ask the other car to swerve -- the goal is to know that you can count on it to swerve if it sees you swerving. (Not that it would hurt to tell it by radio as well.)
This is what fish do, of course. They know the other fish will all turn with them. For them its instinctive, and the trust comes from simply knowing it is your brother fish next to you. For robocars, trust can be established with reputation and vendor certificates if need be. If the car next to you doesn't enter a trust agreement, just give them a wider berth.
For pedestrians, the problem is more difficult. In theory a pedestrian could step into the road at any time. (In California law, jaywalking is illegal and technically, if any pedestrian steps into the road -- not just at a crosswalk -- all cars are legally required to stop. This law is effectively never obeyed or enforced.
But by a computer's standards, pedestrians are slow. And as long as we don't pack traffic ridiculously tight, the cars, like a school of fish, will leave enough gaps so they can part around the pedestrian like fish around a diver.
Then, the skeptics might agree to let the cars on the road.
I've written a more detailed analysis of the school of fish test.
Watching Human Drivers
As robocar technology gets better, it will be installed into cars that are still driven by human drivers. Tens of thousands, perhaps hundreds of thousands of such cars. (Or at least, the sensor tech will be installed, some of it already present for accident-avoidance.)
The human driving activity and all data from sensors will be recorded to build a constantly expanding, giant suite of test data for robocar systems. The robocar brains will play the sensor data and make decisions, and those decisions will be compared to what human drivers actually did. Where the robocars made bad decisions, they will be fixed, where they made better decisions the human data will be corrected or edited out.
Soon each robocar system will be able to be tested as though live through billions of real world miles, including lots of unusual incidents, accidents, all sorts of weather and traffic trouble. And the test set will keep growing and growing (since it is wrong to simply build a system to pass a fixed test set, it must always be tested on brand new test sets.)
Fortunately, human drivers will, every day provide vast amounts of new test data about real world situations and what real human drivers did.
Human Drivers Watching
Once confidence is higher, the tables will turn. The robocar brain will get control of a car, but a human will watch and be ready to correct any mistakes it makes. All this data will be added to the test set.
Soon, we'll be able to evaluate robocar systems and know they can drive billions of miles of real world situations without error. That won't mean they are perfect, but we will be ready to trust them at the wheel without carefully watching them.
As time goes on, the test set will just get larger and larger. When any robocar system does make a mistake noticed by a human, all the different robocar systems should get a chance to check how they handle this situation, so each mistake is made only once. Each new updated version of software will have driven billions of recordings of real world miles.
Special Robocar Highway Lanes
While I don't believe it is practical other than as a temporary measure, we might see the development of special lanes for Robocars on highways, similar to the carpool lanes we see today. And indeed, in San Diego, a test lane of this form was created.
Some expect these lanes to have special markers for robocars, or wires buried in the road that can be followed. I doubt this is necessary. Ordinary, decently marked lanes are enough even for today's technology.
Rather, these lanes might forbid human driven vehicles while robocars are operating. This could allow the formation of trains, which are almost as good as carpooling when it comes to fuel efficiency.
Like carpool lanes in some areas (like L.A.) they might be physically isolated from the human driver lanes, just to keep people comfortable in the early days.
Certainly if such lanes were built, wealthier people would rush to get cars which can use them during commute hours, both for the faster commute, and for the ability to read or work during the commute. The main barrier is how expensive it is to create dedicated lanes.
Robocars Allowed on all Highway Lanes
Once the public is used to safe operation in dedicated lanes, the door will be open for robocars to operate in the other lanes, with human drivers. Highway driving is a pretty easy problem compared to city driving. Like a cruise control, people would bring their cars to the highway and activate the "autopilot." The robocars would probably search around for other robocars and convoy with them, if not form actual trains.
Another early robocar might be the "sleeper car." This would be a car with beds rather than seats, designed for long-haul, overnight traffic at slow speeds. As noted, slow speeds provide much better fuel efficiency.
Users of sleeper cars might pile the family into the sleeper car in the evening and wake up Saturday morning at their weekend vacation home. Instead of leaving at 4pm on Sunday they would leave at 11pm, and plan to arrive at their regular home at 7am with a full night of sleep.
Such cars would have to be quiet with a very soft ride, and not all could sleep in them. I myself have never been good at sleeping on transportation. But some people do it just fine. As seat belts are difficult when lying flat, other safety systems would be needed. (The walls of the vehicle, with less need for windows, could be already inflated airbags.)
Sleeper cars could in theory be used for commutes, allowing much longer commutes by those willing to sleep less at home and catch up on sleep on the road.
Robocars Allowed on Street
Finally, the big day will come in some city, somewhere in the world that's not a war zone. Robocars will be allowed on some urban streets. At first, they will go there under the watchful eye of a human driver ready to take the controls at any moment. But in time, if the designers do their job right, the people will find they don't ever have to take the controls. Then the future of the urban robocar can begin.
This may happen with just a few streets, as a pilot project. Or the robocars might get access first to rights-of-way set apart for Bus Rapid Transit or street cars. Possibly just special lanes or slower streets. The humans may have to alternate between taking the controls and yielding them. At first they will have controls, and require a licenced driver.
For a while.
This opens up many interesting possibilities in vehicle design, and of course means more energy efficient vehicles. Single person pod cars might be more like a fiberglass enclosed electric tricycle than an old-style car.
These vehicles can be cheap and super efficient not just because they are lighter, but also because they don't need much much range, acceleration or speed -- they are just used for short urban trips, which is to say the majority of trips.
To the right you see the Aero-Rider, which is an electric tricycle available today. It uses around 300 BTUs/passenger-mile -- over 10 times better than a car, 3-4x better than the best transit systems.
Note that current ultra-lightweight designs are not sufficiently stable at higher speeds, though lowering the passenger, putting battery weight low down and computer control able to move the battery weight around could solve this.
Once robocars can carry a passenger who can't drive, the robotaxi revolution can begin in earnest.
The anti-human-driver lawsuit
At around this time, I expect we'll start seeing claims (in lawsuits over human caused car accidents) that the human was particularly negligent because they decided to drive rather than use a robocar.
This may be a follow-on from earlier lawsuits that will relate to crash-avoidance systems in cars. The lack of such systems will be blamed for some accidents. The courts will need to decide what that means.
Robocar Only Streets
As the number of robocars increases, and their ability to move more traffic with less congestion becomes clear, we'll see a push for some streets to be devoted only to robocars, at least during certain times of the day, such as commute hours.
Some wonder if, once the world is dominated by robocars, we might see a return to some form of rail for use by vehicles that can run on both regular roads and this rail. If this rail form is much cheaper to produce (including the extra cost of having vehicles able to use it) one could see new roads (especially above or below ground) using this for a smooth, cheap ride. In this case the rails aren't there to avoid having a complex computer driver, they are there because they are cheaper to lay than a whole roadbed.
Robocar Only Zones
Just as many downtowns are considering or implementing "congestion charging" to keep down traffic during the day, we will probably see a move to making some downtowns and other congestion areas robocar-only. Human drivers would need a special permit to enter the zone. People could drive old cars to the boundary, and then catch a robotaxi.
Robocar Only Cities
Some years later, I expect entire cities to ban human driving, or at least strongly discourage it. Human driving will probably still be allowed on rural roads and some highway lines for some time to come to satisfy those who like sport driving. There should also be private tracks for sport driving.
A likely predicted path
This roadmap sounds great, but there are a number of barriers which could stop it. Read about those in The Roadblocks on the path to robocars.