A Futurist Take On Personal Transportation
We’ve had a few posts on this site addressing the coming of the self-driving car. We all know it’s on the way and that many of us will likely live to see the first or second generations of the technology become widely available (if not further generations), so I thought it would be fun to have an omnibus post on the State of the Art, what is on the horizon, what we can reasonably expect to come over the horizon, and what I think the technology will eventually settle at.
Before I begin,I want to say that while I will strive to ensure the information I present here is as correct and comprehensive as I can, I’m going to be covering a lot of bases, many of them outside of my field. I want to take a moment to encourage anyone who has specialized knowledge in any specific area that I touch upon to offer corrections, expansions, or clarifications in the comments.
So, let’s begin!
The State of the Art
One of the hardest things to do is to get a tall, thin cylinder to land on one end. Doing it while blasting an inferno out of one end is a hell of a lot tricker to do, and making it happen from a starting position many miles up in the sky, in freefall, with Gaia knows what kinds of crosswinds is… well… pardon my French, but it’s a salope sans coeur.
But we did it. We taught a rocket how to land on it’s end, safely, gently enough that it can be reused. That’s a hell of thing. We should be proud. We are proud. But it’s also a pretty good indicator of things to come.
It’s very easy to teach a machine how to do a very specific task. The simplest machines don’t even need to be taught, they just do. Think of screws, or levers, or wheels, or complex combinations of them all. The cars I grew up working on had nothing under the hood that I would call even approaching intelligence (and damn little behind the wheel). Brakes worked on direct mechanical connection, or hydraulic systems, same with clutches, throttles, steering, etc. Even the radio was dumb. The only guiding intelligence was human. Not so anymore. Modern cars are loaded with intelligence, and not just rudimentary intelligence. My Tribeca learns my driving habits and adjusts the transmission shifting behavior accordingly. If my wife drives it a few times, it can throw off that education and make the transmission act funny until I reset the memory.
Today, cars are fitted with intelligent cruise control, Anti-Lock Braking Systems (ABS), All Wheel Drive (AWD), Traction Control Systems (TCS), and Electronic Stability Control Systems (ESC). Newer cars have collision avoidance systems as well as automatic parking routines (handy for parallel parking!). And that is ignoring all the navigation aids and other electronic whiz-bangs in the passenger cabin. Hell, I have a dongle that I plug into a port under the steering wheel that let’s me talk to the car’s computer through a bluetooth connection with my phone. All of these systems are marketed as safety and convenience features, as they can detect a problem and react to it many, many times faster than we can, and because these systems have such high sample rates (how often they update the information gleaned from their inputs), they are able to more effectively adjust to rapidly changing situations. Some of these systems are so good that getting a car to actually lose control requires considerable effort on the part of the driver (I have a hell of a time getting my 2008 Subaru Tribeca to slide on packed snow and ice, even with the ESC system turned off, and I grew up in the land of ice and snow; spinning donuts in an empty snowy parking lot is a favorite past time).
They also all form the base technologies needed to have a car that can safely drive itself. That leads to a good questions: what does a car need to drive itself? Let’s look at all the pieces and how they are evolving.
Strip away all the electronics, and what is a car? It’s a chassis that supports a frame with some doors and panels. That chassis also supports the seats, the powerplant, and the wheels. Pretty simple so far, right? Of course, we need to have a drive train, so the powerplant can transfer power to the drive wheels. And we need fuel for the powerplant, so a storage tank and fuel lines, air intake systems, exhaust systems, heat management systems, and a throttle to control it all. And we need to be able to point the car in different directions, so it needs a control yoke and a way for that yoke to manipulate the steering wheels. And that opens a whole bunch of other things that you need, like ball joints and control arms. Of course, roads aren’t perfectly smooth, and even if they were, maneuvering a car requires that the wheels be able to move up and down, so now you have a suspension system. Then you need to be able to stop, so braking system. Finally, there are all the physical system connections (fluid lines, control linkage pass throughs from the passenger cabin to the mechanical systems, etc.).
That’s a lot, and it’s just the basic car. Part of what makes the self driving car a challenge is current car design. None of those systems were ever designed with the idea that someday they’d have to be manipulated by something other than a human. All those baseline systems I talked about above represent significant changes to the internal combustion car model. The sensor suites that allow those systems to gather data, as well as the automatic controls that manipulate the car’s brakes, throttle, and steering have been kinda tacked on, without the basic overall design changing much. Ideally, the self driving car should be redesigned from the ground up, but when you have considerable capital investment in equipment and logistics that support the existing model, starting at the ground floor is a significant undertaking.
So the initial work will happen on normal cars, as imperfect as that is. But what will it look like in the future?
Well, first off, the internal combustion (IC) engine will, at some point, become a thing of hobbyists and luddites. Electric drives are the future. They are simpler, much more input responsive, more efficient, and as time goes on, they will be much more attractive in a power to weight comparison (honestly, we may be at that point already). At some point soon, fuel cell and/or battery technology will cross a threshold and be able to compete with the price, range, and performance of the IC engine, and you’ll see the gas and diesel engines begin to fall away. Cars and motorcycles will go first, since they have the lowest need for such engines, with the higher torque vehicles (trucks, tractors, etc.) following suit as the technology overtakes their performance envelopes. Shipping and aircraft will likely be the last holdouts for internal combustion powered vehicles (and shipping is already using electric motors, they just have IC powered generators).
Your future car will very likely be sitting atop something very much like this.
It may not look like much, but right there is the chassis, the powerplant , the wheels, the suspension, and the steering, all neatly contained in a singular package. There is no linkage for a steering wheel, or throttle, or brakes, because it’s fly-by-wire, so it comes designed already to be self-driving, it just needs a brain to do it. To make it a car, you simply mount a passenger and/or cargo compartment on the chassis with quick connects, plug in a few wires, and away you go.
Of course, it doesn’t have a brain, or any way to asses the outside world, so next we need sensors.
How would a car “see” the road and traffic around it? Lots of ways, actually. GPS is the obvious first way to gather information about the world, as most cars already have that built in. But let’s start with the ways it won’t “see”. Referring back to those automatic control systems up above, most of those operate by “feel”. Wheel sensors keep an eye on wheel rotation speed, combined with inputs from the steering yoke and the speedometer, the car can pretty much constantly know exactly how fast any given wheel should be turning within a very narrow margin of error. If a wheel begins to turn faster or slower than it should, the system will know that wheel has either lost traction, or has suffered a mechanical failure. Together with data from the other wheels, and possibly some accelerometers placed about the car, the system can then apply braking or power to the appropriate wheels to bring the car back under control and, in the event of mechanical breakdown, to a safe stop. This would be even more effective in an electric vehicle with each wheel having it’s own drive motor. 1 Thus a car would “feel” the road, but how would it “see”?
Most likely through a combination of means. The most obvious would be a simple application of RADAR 2 and/or LIDAR 3, possibly with ultrasonic (US) sensors. It could also use visible light, infrared (IR), or ultraviolet (UV) sensors as well, but I think the first three would essentially allow a car to see in all visibility conditions save, possibly, white out blizzard conditions or torrential rain (rain and snow can be damned reflective). So you could also mount the last three systems just to cover all the bases. It may seem like a lot, and a huge cost, but the reality is, every single one of these systems is, or will be very soon, available in a sensor-on-a-chip format.
What is a sensor-on-a-chip format? It’s a relatively new way of producing small, cheap sensors that has really taken off in the past few years. It allows for a low power sensor to be placed on a small computer chip and then easily integrated into a computer for data processing. And I mean small, like the size of a quarter or smaller. You could cover a car in such sensors and the weight of the connecting wires would easily weigh more than the sensors themselves (assuming you didn’t use wireless data transmission). So you could place a small sensor node at each corner of a car and have 360 degree coverage of the surrounding space. Replacing a chip would be like replacing a burned out bulb, relatively quick and easy, and the car would know when it had a blind spot, so it would instantly inform you of a bad sensor and which one.
Finally, a car would be able to hear, sort of. One of the current challenges isn’t keeping the wheels on the road, or avoiding objects, it’s lane following. Once a car hits the road, can it keep to its lane during turns, and can it safely execute a lane change? We actually have a current scheme that we know works very well to keep a robot on track. It’s called a Line Following Robot. It’s a child’s toy, and I don’t mean it’s a toy a child plays with, I mean it’s a toy children can build. And what does every road and highway have painted all along it’s length? Why, highly visible lines! Of course, in a world of self driving cars, road crews would have to be extra vigilant about keeping those lines visible, and there would be an issue in snowy weather with the lines not being visible. So how do we do lane keeping without spending a fortune painting lines constantly and keeping roads perfectly clear? I suspect RFID 4 chips will be a possible answer, as they are cheap, easy to make, rather durable, and quite versatile. If I was to line a road with RFID chips every few feet, basically using them as a radio version of the dashed lines we see, a car could follow them simply by constantly triangulating position with them as a reference. Ultimately, I would imagine such chips would be actually embedded in the surface of the road, but initially, I could see them placed underneath the highway reflectors or under the road striping paint itself. So line following with RFID triangulation as the backup, and the car can keep to its lane, and execute a lane change. This isn’t the only way a car would “hear”, but we’ll get to that in a bit.
Here is where we really start hitting the current technology limits – software. This is why I linked to the story about the Falcon rocket landing at the top, because the software has come a long way, but it still has a bit to go. Vikram (I think, I can’t find the post now) linked to a story about George Hotz taking on Tesla, and how he has a computer that he is teaching to drive. Here at the beginning, this is actually the correct way to do this. Perhaps in the future (that I will touch on later) the programming of a car will be much more straightforward, but as long as self driving cars have to share the road with human drivers, we will have to teach cars to drive like us.
And I truly mean that, we will be teaching them. Machine learning is a thing, we know how to do it. Perhaps not as well as we’d like, but we are getting there. This is largely what Tesla, and Google, and George Hotz, and quite a few other companies are working on, teaching a car how to drive amongst humans. It’s a challenge, but a surmountable one. Give it a couple of years and we’ll see some pretty significant breakthroughs on this front.
One way to move it along is to have cars talk to (and “hear”) each other. In concept, this is pretty straightforward. Each car broadcasts a lower power radio signal with key operational information such as a unique ID number, speed, and any expected immediate changes (“I will be moving one lane to the right in half a mile”, “I will be slowing to 45 MPH in 1 mile”, or “I will be turning left at the next intersection”). Nearby cars will hear that information, then be able to react accordingly to allow for the coming course correction. Such information would cascade backwards in a way that will allow for traffic to efficiently allow the car that needs to move to do so. Such information is actually available today via brake lights and turn signals, but the information is terribly limited and its transmission is very inefficient. Unfortunately, the current average age of cars on the road is about 11 years old, so it’ll easily be a decade before self driving cars could hope to even begin to achieve a majority presence on the road where such broadcasting could have a noticeable effect. I suppose we could devise some manner of aftermarket device that will tie into an older vehicle’s navigational system and serve the same function, but getting people to adopt such a device will probably be inconsistent enough at the start to sill be an issue. So smart enough to deal with irrational humans it is, at least for the immediate future.
The More Distant Future
So where will this all go? I expect, as the necessary engineering and software hurdles are crossed, you’ll first see self-driving cars mostly as” Advanced Adaptive Cruise Control” (get on the highway, turn it on, and relax until it’s time to leave the highway – truck drivers will love this!) and “Advanced Parking Assist”. This mode of travel will persist while they work out the kinks and then it will begin to work on major arterials, then side streets, and finally you’ll hit truly autonomous driving. Along the way, I won’t be surprised to see cars that can interface with a phone, such that as I leave my desk at work, I could send a message to my car and have it meet me at the front door, instead of having to walk through a parking lot or garage.
Once we hit truly autonomous, I expect the market for personally owned vehicles will contract, a lot, especially in urban areas, and the taxi industry will very quickly die (or, at least, taxi drivers will become a thing of the past, taxi companies will probably be the first to buy and deploy such fleets). Owning your own car will be a luxury or a geographic necessity. Cities or companies will own and maintain fleets of autocars that can be dispatched by app for moderate rates (including ride sharing to save costs). The chassis system I showed up above, and systems like it, will be the backbone of such fleets 5. Long range vehicles will likely have a slightly different build that includes a standardized swappable battery (gas stations will become charging and battery swap stations). I wouldn’t even be surprised to see cabin roofs fitted with some manner of photovoltaic solar cells, so a car sitting in the sun can charge itself.
I also expect the roads will get a lot dimmer. Cars won’t need nearly as much light as our eyes to “see”, so headlights that can illuminate a road far enough ahead to drive safely at 60 mph at night will be gone, and cars will just have LED indicator lights so they can be seen at night, not so they can see.
Eventually, human guided vehicles will be the domain of motor sports and hobbyists, and I would not be surprised to see legislation change such that human piloted vehicles will have to have a communication system tied to a nav system (so autocars can “hear” them), and they’ll have yield the right of way to autocars whenever the right of way is in doubt.
The largest problems I see in this new paradigm is that of security. If a car can be dispatched, then it has to be able to accept commands from a central dispatching station. If it can accept commands, it can be remotely hacked. I suspect that even personally owned autocars will have to be able to accept remote commands at the demand of law enforcement, so it won’t just be fleet cars that can be hacked. How autocar security will be implemented will largely depend upon the system architecture, and current car makers have to take seriously the hard lessons about how to design such systems, or I foresee them suffering a lot of bad PR and lawsuits.
Still, if the security issue can be adequately addressed, I expect by the time my son is 16 (he’s 3 now), getting a driver’s license will be a rite of passage that means very little to him. And this will probably be a good thing.
- In previous discussions, there have been concerns that an autocar would be unable to handle bad weather. I’m afraid that is the wrong question to ask. The right question is, “How is it you haven’t died yet because of bad weather driving?” Cars lose control on the road because they lose traction. Usually this is due to some physical phase of water on the road causing a loss of traction, but it could also be an oil/chemical spill, sand, gravel, or a loss due to too much speed. People wreck because they are unable to restore traction in time, which is actually pretty understandable, since we have two strike against us when it comes to regaining control. First, we are terribly ignorant of what our wheels are doing. By the time you feel the changes in your hands, in the sound, in your inner ear) that will indicate to you that you are losing control, the car has already started to slip and build up momentum in a bad way. Strike One. Then your brain has to recognize you’ve lost traction, and then it has to decide what to do, all the while, the car continues to build up momentum. Strike Two. Professional drivers, who have experienced hundreds of slides/skids, gain an advantage in that they can recognize and respond to a loss of traction without thinking about it, but most people just can’t. The system described above can recognize the danger and take corrective action a hundred different times before the driver would even have enough information to recognize the problem.
- Radio Detection and Ranging
- Light Detection and Ranging
- Radio Frequency Identification
- If each chassis can accept a variety of passenger/cargo cabins, a fleet would maintain a number of chassis and a slightly larger number of cabins of different types, such that a standard passenger car could quickly be reconfigured into a light van or truck, instead of having a selection of light trucks sitting around just in case. This assumption is based upon the idea that the bulk of the cost will be in the chassis because most of the cabins will be somewhat spartan and inexpensive (think about the interior of a city bus as compared to a tour bus – durable and easily cleanable as opposed to exceptionally comfortable), while an option for a more well appointed cabin will be possible (for an increased cost). Keep in mind that the bulk of the cost of the cabin of current cars is in creature comforts (because it is your car) and all the system pass throughs for controls and HVAC. If the cabin is just a box that sits atop the chassis, the cabin is a lot easier to design and cheaper to build