Replies

I’ve also spent 35 years+ in software, still am. While I hear, very loudly, what you’re saying I’ll point out that 99% of the time airplanes are flying themselves. I work in automotive mainly, meeting functional safety development standards is hard, it’s not like developing code in most environments. Ultimately, every function must trace to requirements, every requirement must trace to code, and one or more tests must exist tracing back to every requirement. Static analysis tools *must* be used, MISRA-C coding guidelines followed, as well as SIL, PIL, HIL, etc. testing performed etc. etc. Then there’s meeting SOTIF standards on top.

IMHO - it’s knowing this that is causing such headaches. It’s orders of magnitude more expensive to certify code, as required, than regular software. For this application there so much code (tens of millions of lines) that this standard approach, as followed in avionics, is an economic problem.

From my view, the industry is reverting to long laundry lists of ADAS capabilities - I have a list of 30 that are going into vehicles. I do believe autonomous vehicles are inevitable but the problem is harder than most thought. I’d rather the investment be put into technologies that, in the short term, save lives - this should have been the focus. This is now the trend - but full autonomy distracted from this goal, even if a lot of derivative work does go into ADAS features.

Fellow code jokey here. Our new car has self-driving features (they call it "driving assist" with the word "assist" meaning we're just helping you if we see you drift out of the lane, but don't depend on us and sue us). I played with it on quiet roads and disabled it the first day. If I veered some onto the center lane to avoid someone checking her mailbox, it tried to "help" me put it back in the center of the lane (too close to the pedestrian). Same with if I veered the other direction to avoid a large truck that veered into the center (for a sharp curve). Either the software is buggy or it's unreasonable to expect the design to account for so many variables.

I would imagine autopiloting a plane doesn't have to account for so much other random traffic.

I’ll point out that 99% of the time airplanes are flying themselves

Connected to TCAS, separated by a thousand feet and miles by design and flight plan, and listening to enroute controllers. "The crowded skies" aren't.

I've been flying for longer than I was working, and I know first hand that the only place it gets crowded up there is in class "B" airspace, and around airports - where autopilots are disengaged.

Glad to hear, though, that automotive companies are taking the danger of their autonomous activities to heart and increasing the rigor through which software is certified - but that reassures me not at all.

While I hear, very loudly, what you’re saying I’ll point out that 99% of the time airplanes are flying themselves.

Great point. I'd offer two observations related to this:

1. Airplane manufacturing is probably one of the most heavily subsidized industries in the world. Much of the R&D in the industry is done under military contracts, so the industry does not have to pass the costs of these things along the way the auto industry does.

2. When it comes to technology development, one of the biggest limitations of the auto industry is that there are very different economies of scale when it comes to building costs into the price of the product that is sold to the customer. Incorporating technology in a car that is: (A) rarely used for more than several hours in a typical day, and (B) is driven by a single occupant most of the time it is being used, tends to be very costly to the end user -- especially in comparison to aircraft that carry hundreds of passengers and is kept in service for as many hours as possible.

Ever notice how there are far fewer pedestrians, or dogs and children running in front of your vehicle, or stop signs, stop lights, and railroad crossings, up in the air as compared to on the road?

Yes, flying has its own unique challenges, and I do not pretend to be able to list all the known factors; and many of them have to do with local weather patterns and the effects on flight, as well as monitoring the physical responses of the plane’s control surfaces, shutting down hot engines, icing, etc., which aren’t as applicable to cars (”Captain! The steering wheel isn’t responding!”)

But from a layperson’s perspective, it looks like a lot of the airplane constraints, revolve around things which were designed by people as part of the plane, and can therefore be quantified within some error bounds, as opposed to the arbitrary and hard-to-identify-and-differentiate-instantly, interruptions (”bouncing ball in the road implies little kid to follow”), which drivers on the road have to deal with.