What will we have first: Flying Cars or Autonomous Cars?

Or, will we ever have any of those 2 things?

Really, chl? Trying to predict the future now?

Well, I think I can answer that because I am a physicist, a car driver, a (lapsed) private pilot, I know computer science a bit and I follow the news about all of this.

Furthermore, we already know that real flying cars are impossible with today’s or even tomorrow’s technology, unless we figure out a simple anti-gravity device or something similarly miraculous. So this reduces the question to “will we ever have good enough AI to have really autonomous cars.”

In the rest of the article I give some technical details about flying cars, formulate some critique of AI/deep learning, and then I move the goal posts until I get different answers.

Regarding the infeasibility of the real flying car, one needs only look at a typical car: A >1.5 ton behemoth engineered to go fast on roads and to keep its squishy payload alive in the frequent crashes that happen on our roads. Then compare this to a typical 4-seat small airplane or helicopter. Piston engine driven small planes are always marginal, to the point that filling all seats with passengers often means being limited to half-full tanks, else you would be overweight. This means that they are flimsy (compared to cars). A two-seater like the Cessna 152 weighs less than 1 ton empty, and that includes all the flying stuff like wings and propeller. (It will also be 40 years old, yet still be reasonable to fly. That’s how different the economics of flying are compared to driving.)

The less marginal plane models cost upward of half a million euros and are still not as crash-worthy as a car. Because they do not need to be. Flying is not like driving, and many kinds of crashes are not survivable anyway. The trick is not to crash, or at least to hit the ground at a glancing angle. Which leads us to the topic of pilot certification. Most of the training is geared towards avoiding flight into bad weather, flight with too much weight or too little fuel, and there is radar tracking and air traffic control. There are stringent planning requirements for each flight (weather, weight and balance, NOtices To AirMen, flight planning (navigation) etc.) So, right off the bat, even if you could get a roadworthy vehicle into the air, the flying portion would have to under tightly controlled autopilot procedures. Ironically, autonomous flying in today’s technologically advanced air spaces is a solved problem, at least the technological part is (if you only want to go from specially equipped airport to specially equipped airport in reasonably safe weather conditions).

So, even if you slimmed down the car, you would still be adding heavy wheels and suspension to a plane that would then be overweight. So you need a more powerful flying engine. A typical traditional plane engine is an air cooled 4 to 6 cylinder boxer that develops peak power at 2000 to 2400 rpm, directly driving the propeller without any gearbox. 110 horse power is enough for the aforementioned Cessna 152, which has a displacement of 3.3 liters. Cars get 120 hp from 0.99 litres, albeit at 5000 rpm. So, do we need an additional driving engine? And yes, automotive engine conversions for use in planes exist, but not in the high power range we would need for a flying car. Another delightful detail of aviation: You need a new or overhauled 20000++ € engine every 2000 flight hours. And the driving would count toward that limit. And you will have to stow the wing in a useful manner during driving.

Then what about helicopters? The thin main rotor is more easily stowed and the ability to hover makes them much easier to guide in congested airspace. Unfortunately, helicopters are even more marginal, because they figuratively have to beat gravity into submission to get airborne. A helicopter with more payload than 2 medium-sized adults plus 1 handbag will have a turbine engine. Lots of power, lots of fuel consumption, hilariously high operating costs, and not really suited for driving wheels.

In conclusion, the wildly diverging engineering demands on cars vs. aircraft make the flying car a la Jetsons impossible for now. This kind of duo-modal transport will still entail vehicle changes in the future.

So far I have talked about real flying cars. What about unreal ones? For this, we need to move the goal posts. Let’s drop the real car requirement and take something like a dune buggy. Let’s also get rid of the huge rigid wing that is such a PITA on the road and use a fabric wing like a para glider. Such a thing exists as a prototype, but you still need to be a pilot. Or, take an autogyro, whose 2 thin main rotor blades are easily folded and secured. Neither of these are “proper” planes, in Germany they would be in the “air sport devices” category. A totally different take would be the personal air taxi Volocopter. It is basically a very large drone with multiply redundant rotors and battery systems. Made for semi-autonomous city flying. And with no wheels at all.

Regarding the autonomous car: This section is short, in proportion to my expertise in the matter. A really autonomous human-like driver in general traffic needs a very powerful AI. Some companies use superhuman sensors (LIDAR + cameras etc.), others want to get all information just from camera images. If you use deep learning for this, you have the problem that you do not really know how the computer is doing it. And if it will be hilariously wrong in some situation in the future. A simple example includes deep learning to identify dogs in pictures. Now, huskies happened to always be shown in snow in the training data, so guess if the husky in the living room was classified correctly. Will the humans responsible for the autonomous cars be able to rule out all the blatant and the subtle issues with the training?

Moving the goal posts: What I think could happen, barring any world shaking breakthroughs in AI: There could be special, possibly fenced off lanes for semi-autonomous vehicles, no pedestrians, all vehicles must participate in central or cooperative control. Would such a system solve a traffic problem that we have? A traffic expert needs to answer that.

To sum it all up: Already lots of known physics and engineering knowledge makes a 1950s sci-fi flying car unlikely; the autonomous car has more wildcards, I personally think (guess) we, humanity, will not create a good enough AI for a general robo-car like in 1950s sci-fi.