Hacker News new | ask | show | jobs
by rbanffy 1382 days ago
I wonder if there isn't an easy way to automate the parabolic flight - the autopilot should be clever enough to do it by itself without needing human intervention.
5 comments

The real question is "does it cost more to certify the autopilot to do this than it does to pay for an extra two pilots", and with my understanding of aerospace costs that answer would be "very, very much yes".
I could imagine the autopilot could do a much better job.

In the ideal 'zero-g' flight, you can release a ball at the start, and it should stay stationary (relative to the cabin) for the whole flight.

That means the plane body must be to within a few inches of the exact perfect location in the sky 30 seconds later.

I haven't flown planes, but I imagine that making sure they follow precisely a path with an accuracy of an inch in all three dimensions while travelling at 600 mph is no easy feat. But an autopilot with a fast control loop might be able to do it.

All the autopilots in the world aren't keeping a plane to within an inch in all three dimensions. It'd be nice, but turbulence is a thing :)

More importantly, actual position isn't that important - you care about relative acceleration being as close to 0g as you can.

And with the 3-pilot setup the A310 zero g flights have, they maintain +/- 0.02g. Now you add to that the fact that developing an autopilot you can trust in that situation costs a good chunk of money, the fact that it's a specially modified plane (so it's more or less a one-off effort), and the fact that pilot costs are negligible compared to the rest of the flight cost.

At this point, you can get a minimal increase in precision for a large expenditure up front. You'll still have the 3-pilot setup (you want to be able to recover on malfunction), so operational costs aren't reduced.

At that point, the question becomes "why would you"

The “easy” way to do it is to mandate autopilots to have the capability for precise ballistic trajectories. This would need better positioning, accelerometers, and meteorological radars and would also enable fully autonomous landings (if the plane can be aware of air densities and speeds around it, it can nail the landing even with wind gusts).
Wait what? If your landings are a ballistic trajectory, kindly step away from that stick ;)

And that's kind of the point - nobody except the few planes that fly parabolic trajectories needs this. So good luck mandating something affecting all of aviation with no practical use except for a handful of flights.

Autoland already exists. And can handle crosswinds up to 25kts. It's just that using them is an incredibly intense task - your reaction time to fix mistakes is a bit short.

That's why most landings are still manual, unless visibility is so bad that you have no choice.

(Also, the idea that meteorological radar could give you enough info about air density to preemptively handle air gusts is... a bold future)

I didn’t say parabolic landings. I said the same capabilities required to fully automate parabolic flights would allow safer landings because precise trajectory control even when there is turbulence around the plane is a very desirable feature.

On second thought, it could pay for itself in preventing aborted landings.

Most passenger airliners flying today can land automatically, but are only allowed to do so when visibility is so bad that it's the only option. It's called CAT III Autoland.
I guess even a supercomputer would not be able to compensate for wind gust, turbulence, etc.
There isn't much turbulence, wind gusts etc. at altitude - these are more of an effect of the bumpy and uneven, alternative hot and cold surface at ground level - the bumps are usually all smoothed out by the time you are high up in the air.

(besides large storms, of course)

EDIT: Turns out there is sometimes some turbulence at altitude, caused by large mountains, storms and the jet stream - but these are all predictable and avoidable.

> "does it cost more to certify the autopilot to do this than it does to pay for an extra two pilots"

I bet it does. Parabolic flights aren't really a huge market.

a pixhawk, ardupilot, some Lua, and a handful of servos and you're there. ;)
On an A310 you don't even need the servos.

But you probably want better than 8 bits resolution on sensors and output.

"Wow, this is hard. Therefore, AI must be the cheaper solution!"
This doesn’t need AI. The physics are very well understood. All that’s needed is a better autopilot with better awareness of surrounding conditions and more precise control of the aircraft.
It's plain old Control Theory, no AI needed.
It baffles me that they need five pilots to accomplish this.
FTA "To fly a parabola takes 3 pilots working in concert. Two pilots sit in front of the control columns and a third sits in the center jump seat to manage the engines. When flying a parabola, one pilot is responsible for the pitch of the aircraft and the other for the roll. The pilot responsible for the roll of the aircraft actually has a modified control column in order to put as little pressure on the yoke as possible, allowing the pitch pilot to focus on the correct parabola timing." I think precise timing requires many pilots. Lots to keep track of in the name of science.
Yeah doing a parabola is relatively easy. Doing a nearly perfect one to maximize weightlessness without overstressing something is harder, so having one pilot for each aspect makes sense.

Another for radio and another for “general everything watching; the captain” makes sense.

My aviation family has a long history all the way back to training pilots for WWII (already highly experienced - too much so for combat) so my hunch is accuracy and redundancy.

Under normal conditions, sure it’s a one time maneuver, within limits, etc. This is much more of a risk-laden undertaking with the repeated stress on the airframe, the human body, and so having individual attention on important individual tasks - with experience and brain computing speed machines still can’t match for a brilliant stick and yoke pilot in unexpected conditions - this is for safety.

Also considering my background with French culture, I imagine the discussion went something like “we could do it with fewer personnel, but why? This works.”

I imagine it's for an abundance of caution, as they are flying these planes way outside what their normal operations manual recommends. I assume a 320 NEO or a 737 MAX would slap the hand of the pilot if they attempted to go for a +50 degree climb and scream expletives in the voice alert system.
Okay you just gave me a mental image of an Airbus autopilot system so angry at you it starts swearing in French. Totally agree with safety and ‘normal use case this is not’ situation.
Bombardiers will issue a stream of Québec sacres:

https://en.wikipedia.org/wiki/Quebec_French_profanity

It kind of does already: https://youtu.be/vmbzKsqKQoI#t=30s
Defaulting back to it's `native` language (of the programmers). That would be hilarious and should be on the next AirPlane spoof.

Mel Brooks, where are you!

I thought about a cartoon robot articulated arm with a white glove trying to wrestle the control column from the pilot.
Or the autopilot computer ejects by itself with a small parachute. Very Looney Tunes!
"I'm sorry, Dave. I'm afraid I can't do that."
Well...in MS Flight Simulator, I know the voice alerts will yell at you to warn you about terrain, of if your AoA is too great ("Sink rate!").

But they don't care about high climb angles until you eventually stall.

Yeah let's put Musk's autopilot in it and lose 20 people every few years when the plane get scared by the shadow of a cloud. That'll save money
it's a lot more complicated than it sounds:

https://youtu.be/M67YP-f-LyI?t=592

The “vomit comet” (American equivalent of this) has autopilot for yaw and roll only.