Is there an autonomous flying car in your future?


About 100 companies, large and small, are racing to develop certifiable airframes that can handle up to six passengers; some think they will be flying by 2021, most by about 2025, and there are predictions of thousands of such “flying cars” serving as on-demand air taxis, landing at “vertiports” atop tall buildings in megacities around the world.

“If you’re going to do anything interesting with this as a large volume, you’re going to have to get rid of the pilot license,” Dr Luuk van Dijk told ICAO’s Journal magazine. For flying cars to be economically viable to build, he estimates 1500-1700 flight hours per year per aircraft. And since there’s already a shortage of licensed commercial airline pilots (and flight instructors), “You won’t have enough CPLs that want to be a taxi driver.” Van Dijk is the founder and CEO of Daedalean, a 22-person startup in Zurich, Switzerland which is developing autonomous guidance, navigation and control software for small electric personal aircraft.

“These electric vehicles are much simpler than aircraft with combustion engines. And because they’re simpler, they can be cheaper, they can be quieter, they can be safer. There’s no reason a Lilium or Volocopter should cost more than a Tesla Model S. And there’s no reason you shouldn’t be able to sell 50,000 of them,” Van Dijk explained.

The Daedaleanleader, a physicist, has worked at Google and SpaceX (designing spacecraft flight control software), and several years ago led a project designing an autonomous ocean-going sailboat at ETH, Switzerland’s technical university which has produced 21 Nobel Prize winners (including Albert Einstein). Van Dijk said there are “five or six groups” in Zurich “that are at the forefront in robotics and autonomy, flying control theory, deep learning.”

Swiss startup Daedalean has recruited experts in robotics, mechanical engineering, computer vision, artificial intelligence and avionics. (And rescue dog, Daila.)

In 2016 he founded Daedalean. “I wanted to do the most interesting startup I could possibly think of. The startup scene was clearly lacking in ambition, in my view. Startups go for acquisition really early and when they get bought the talented students are shipped off, never to be heard from again. I thought that was a massive waste of opportunity. When hiring you have to compete with big companies that pay very comfortable salaries. So you can only lure people if you have something that is obviously more interesting.”

Daedalean has spoken with 30 to 35 of the electrical vertical take-off and landing (eVTOL) developers. “Some people really know what they’re doing and they’re on a good track and they’re going to be on the market by 2021 – this is the median number being thrown around. Full autonomy is on everybody’s list.” “We start with everything a human pilot does, then we will try to build systems that can outperform a human in every way. So to speak, a drop-in replacement for a human pilot,” described Van Dijk. “Not in the form of an R2D2 that sits in the copilot seat and has a robot arm control of the stick. Electric VTOLs are already fly-by-wire; the primary flight control computer already has sufficient control over the attitude, altitude, trajectory, etc. to fly reliably from A to B.”

The first part of the system that Daedalean is focused on – the Visual (X) – is designed to replace the eyes and the visual cortex of the pilot “to see where we are and what we’re looking at. We’re doing that for the ICAO-defined categories and we’re training neural networks to recognize them.”

Systems using simultaneous location and mapping (SLAM) apply the parallax of onboard cameras “to solve the equations that tell you where you are in space and where all the other things are … more accurate than a GPS.” Van Dijk noted that software for safety-critical applications in aerospace must meet such standards as DO-178C (Software Considerations in Airborne Systems and Equipment Certification), DO-254 (Design Assurance Guidance for Airborne Electronic Hardware) and DO160 (Environmental Conditions and Test Procedures for Airborne Equipment). He said, “We have to convince EASA and the FAA that neural networks are safe enough. We also have to help develop the means of compliance to demonstrate the thing. We intend to be able to demonstrate safety in the way that the authorities are accustomed to, so we need not ask for special exceptions because that will never fly. Fortunately, they are moving to performance-based metrics rather than prescribing how it should be done; the winds seem to be even more friendly.”