Is there an actual reason why the control tower work can't be fully automated? For train control lights we almost don't rely on human operators anywhere.
> For train control lights we almost don't rely on human operators anywhere.
Trains are on tracks. They basically move in one dimension. And the tracks can have (near) contact-based sensors along the way where the exact distance is known. (And in the US, there still is human conducting in a lot of the US)
That’s a very different problem space than the three dimension, unattached, space that air traffic moves in.
Because things can go from routine to multiple simultaneous life-threatening failures very quickly. Something like one flight declaring a mayday while another one just lost communication, all while the radar just started glitching in a weird way. Human intuition and common sense can sort it out. Deterministic algorithms would not.
E.g. On 9/11 ATC had to land almost 3000 planes in 1 hour. I'm not sure if that sort of national coordinated grounding is part of ATC training, but it's certainly not something I'd want to leave to some code that has never needed to run in production before.
It just seems like software could pretty easily compute non-intersecting flight paths for all planes and assign them accordingly. As well it could real time monitor all trajectories and continue to give out the updated flight paths. I don't see why you also couldn't run a trillion tests using real and simulated flight data to make sure it works well.
Air traffic control has a lot more things to deal with. There are scenarios like runway closed and all traffic has to be diverted. Loss of communication. Various emergencies. Weather changes. It's not just a question of 3D motion planning. Controllers in the tower also use their eyes.
In your imaginary system how is the software "tower" communicating with airplanes, using voice? I don't think we even have software that can reliably decode the variety of human voices over radio that a controller can respond to.
One can imagine a digital protocol to all airplanes but technology works its way really slowly into aviation.
Yes, it would require a piece of hardware, but that seems easy to regulate and wouldn't need to be very expensive certainly not in relation to the costs of owning and operating an airplane.
Seems like you’ve stumbled onto a very obvious solution that would be easy to implement and that no one else has ever managed to see, but which would totally revolutionize the airline industry. Time to start a business!
Because right now airplanes are flown by humans. A large part of atcs jobs is dealing with humans. Not every pilot will be able to fly the optimum flight path.
I Googled about this. London Heathrow is widely regarded as the busiest two runway airport in the world. They allow less than 50 arrival per hour. Are there any airports in the world that can have "hundreds of arrivals per hour"? Conservatively, 200 can be considered "hundreds", then you would need 8x runways operating at max capacity. That seems hard to imagine. Google also tells me that Atlanta (normally the business airport in the US) can handle about 250 "operations" per hour, so let's say half for arrivals.
Serious question: Why is 3D such a hard problem for modern computers? I could imagine a plane enters a cylinder of airspace near the airport and automatically communicates by radio waves information about itself. Then, HAL9000 can provide guidance as a landing plan.
I think the 3D part is fine, it's more "all of physics" coming into play with a plane. Like bird strikes, engine failures, trying to decide how to handle that. What is the loss function the AI is supposed to apply if an emergency landing is needed?
And then if you have people flying the planes, you have to deal with people mostly _but not always_ doing the right thing. So now you build out a plan and have to deal with consequences of that.
So at the end of the day you're still looking at funneling humans into a thing. At worst you could consider ATC as "customer support", there to press buttons on machines to actually handle a bunch of logistics because the pilots need to figure things out.
On top of all of this, airports are trying to get through a lot of flights quickly. So people can make snap judgements about whether planes can or cannot advance, what they should do, etc. No matter how well your plan is, the instant a pilot mishears something it's over.
If we can figure out self-driving cars, maybe we can talk about replacing pilots with AIs. But in the meantime there's somebody not following the plan often enough.
This does lead to an interesting question for me, though: what is the biggest "human movement" system that is actually entirely hands-off logistics? I would imagine that postal service companies are doing a lot but every major person moving operation seems pretty hands-on from the outside.
> If we can figure out self-driving cars, maybe we can talk about replacing pilots with AIs.
There's currently a human in the loop at almost all times but a great many planes are already self-flying. Autopilots are a thing, and have been for decades. Modern airliners routinely land themselves, and Garmin have developed an emergency landing system for general aviation aircraft which can handle everything including selecting an airport and communicating with ATC in the case of the pilot being incapacitated.
In many ways a self-piloting plane is an easier problem to solve than self-driving cars. Every plane has (or will have in the near future) a beacon on it transmitting it's location to every other plane around it to allow collision avoidance, and the procedure for getting from one major airport to another is pretty much prescribed in the form of standard departures, arrivals, and airways. The big difference of course is that if a car on the road breaks down and the AI can't handle it then the car can just stop, and baring someone not noticing and plowing into the back of them everyone will be fine, while if that happens in a plane it's a matter of time until the thing falls out of the sky killing everyone on board.
I should have said “arrivals and departures”. KATL has 2,100–2,700 arrivals and departures per day. Even if we assume those are equally spaced throughout the day (they aren’t) we’re over 100/hour.
But sure. It’s a mild exaggeration. It still doesn’t change the core point.
3D isn’t just “magically hard” for computers, but the process for routing traffic is wildly complex and a bunch of planes arriving and leaving at semi-random times, directions, and with different requirements and capabilities is where the problem starts not finishes.
The happy path is relatively easy. The exceptions are innumerable.
It's not just a 3D problem. It's all of the management in the air and on the ground. Could a computer eventually do it autonomously? I would think eventually, but the problem is handling exceptions. The Navy has been experimenting for years with a digital replacement of their Ouija board analog flight deck management tool for carriers [1]. And even then, people are still making most of the decisions.
And ATL is a crazy busy airport (there's an old doc on Netflix I think which is interesting). To confirm your question, ATL can run 5 runways nearly continuously[2]. It would be interesting to know what they peak at during a busy Monday morning, but my guess is they are more constrained by gate space at this point.
> Serious question: Why is 3D such a hard problem for modern computers? I could imagine a plane enters a cylinder of airspace near the airport and automatically communicates by radio waves information about itself. Then, HAL9000 can provide guidance as a landing plan.
Maybe I'm getting older. Maybe it is a more modern phenomena, enabled by a steady diet of 30 second videos.
Either way, I'm quite disturbed, regularly, recently, by the # of people who breezily stumble through quarter-baked thoughts while speaking dismissively, as if they've covered the surface of a complex universe that has been worked on very many smart people for decades, and now we can get to the real singular problem that'd fix everything, the one thing they've identified.
I don't even know where to begin trying to interlocute when the starting premise is "3D is hard for modern computers."
So I speak straightforwardly, in a way that I wish wouldn't be seen as rude, but it is.
So it goes.
You're right about the arrivals, but you missed the forest for the trees in the comment you're replying to.
"It" is an indefinite pronoun, meaning, "it" is by definition unclear to anyone but the speaker.
Here, "it" means "your involvement in this discussion"
To put it a bit more plainly, and apologies if it is hard to hear currently, I'd find this very helpful to hear this if I were you:
- The first comment was objectively, straightforwardly, dismissive. You questioned one statistic, then ignored the thrust of the comment and the comments leading up to it, instead framing the difficulty as "modern computers have a hard time with 3D", eliding the main point communicated.
- You then replied to my comment to tell me I'm wrong, it wasn't dismissive. This is also, in itself, dismissive. I had already fisked you, and I felt bad, so I didn't want to get overly literal and criticize you roughly again, in front of a crowd, so soon. So, I choose to keep it brief, rely on your ability to recognize the irony in your reply being dismissive in telling me the original comment wasn't dismissive, especially when combined with the wink.
To be hyper clear, it's...unusual...when communicating with others to ignore most of what they just said and fixate on one illustrative part of what they said, then when told you were being dismissive, to just say "I wasn't being dismissive" and put the burden on the listener. It's so unusual as to be objectively amusing.
You certainly did not mean to be dismissive, but you were. I sympathize with the reflex, I am 100% sure I've made many similar errors in my own journey, probably more than you ever have and will. It is a problem area for me.
With humility, I'd humbly suggest that when given feedback, rather than briefly asserting the feedback is incorrect, you come with curiosity and ask why it came off that way.
Given their answer, you'll be able to tell very quickly if the person is having a bad day/picking on you, or if you were unclear.
I'm a different person from the guy you originally called dismissive.
Your advice would make sense if I was the same person, but if I was the same person I would have responded very differently to the criticism. I gave a brief reply because I'm a third party.
Four Dimensional Navigation, actually! X, Y, Z, plus Time. By incorporating strict time constraints, air traffic controllers can schedule and merge arriving aircraft more precisely, reducing holding patterns and optimizing fuel usage.
>4D AREA NAVIGATION SYSTEM DESCRIPTION AND FLIGHT TEST RESULTS
>A 4D area navigation system was designed to guide aircraft along a prespecified flight path
(reference path) such that the aircraft would arrive at the approach gate at a time specified by
the ATC controller. Key components to achieve this requirement were:
>(1) stored reference trajectories;
>(2) a continuously recomputed capture trajectory to a selected waypoint on the reference trajectory so as to achieve the desired time of arrival;
>(3) electronic situation displays; and (4) a control system to follow the overall trajectory in space and time.
>Four-dimensional guidance algorithms for aircraft in an air traffic control environment
>Theoretical development and computer implementation of three guidance algorithms are presented. From a small set of input parameters the algorithms generate the ground track, altitude profile, and speed profile required to implement an experimental 4-D guidance system. Given a sequence of waypoints that define a nominal flight path, the first algorithm generates a realistic, flyable ground track consisting of a sequence of straight line segments and circular arcs. Each circular turn is constrained by the minimum turning radius of the aircraft. The ground track and the specified waypoint altitudes are used as inputs to the second algorithm which generates the altitude profile. The altitude profile consists of piecewise constant flight path angle segments, each segment lying within specified upper and lower bounds. The third algorithm generates a feasible speed profile subject to constraints on the rate of change in speed, permissible speed ranges, and effects of wind. Flight path parameters are then combined into a chronological sequence to form the 4-D guidance vectors. These vectors can be used to drive the autopilot/autothrottle of the aircraft so that a 4-D flight path could be tracked completely automatically; or these vectors may be used to drive the flight director and other cockpit displays, thereby enabling the pilot to track a 4-D flight path manually.
>4D-TBO: a new approach to aircraft trajectory prediction
>How four-dimensional trajectory data could contribute to aviation decarbonisation targets
>The real-time transmission of four-dimensional trajectory data has the incredible potential to greatly improve an aircraft’s trajectory prediction. By reducing the inaccuracy of current air traffic management (ATM) prediction models by approximately 30-40%, the Trajectory Based Operations in 4 Dimensions (4D-TBO) project is helping to pave the way to a more sustainable management of tomorrow’s air traffic.
>The 4D trajectory of an aircraft consists of the three spatial dimensions plus time as a fourth dimension. This means that any delay is in fact a distortion of the trajectory as much as a level change or a change of the horizontal position. Tactical interventions by air traffic controllers rarely take into account the effect on the trajectory as a whole due to the relatively short look-ahead time (in the order of 20 minutes or so).
>The implementation of 4D trajectory management is being researched by SESAR (Single European Sky ATM Research) in the EU and NextGen in the US.
>The 4D trajectory concept is based on the integration of time into the 3D aircraft trajectory. It aims to ensure flight on a practically unrestricted, optimum trajectory for as long as possible in exchange for the aircraft being obliged to meet very accurately an arrival time over a designated point.
Yes, this multiplies the complexity. When you talk to ATC you always need your tail number and airplane model. Why? Because a landing Cessna 150 is moving at 70mph. An incoming jet is moving at 130mph. And the jet can’t just slow down to 70 or it will fall out of the sky. They need to consider aircraft performance in all aspects of planning.
How would you deal with all sorts of emergencies involving human pilots? For unmanned aircraft(aka drones) it’s a lot easier to implement unmanned traffic management (UTM).
Direct them to land in the best location and make sure all other aircraft are on non-intersecting flight paths. What exactly is ATC doing that software isn't able to?
How many options are there for handling emergencies with aircraft now? You pretty much just have either land ASAP or circle to burn fuel and then land.
I mean this sincerely. What is more likely: that we've spent several decades ignoring very real automation solutions to this problem, or that it's a really, really hard problem that could get people killed?
Not an expert but I can't imagine that would go very well. Trains have a single axis of movement that sometimes cross or combines with others. Aircraft have three axes of movement all under human control.
not a bad question honestly.. I'd want some highly skilled humans there monitoring things but, yes. The air traffic control system of the US is absolutely incredibly amazing, but their entire mission, technology and equipment used to accomplish that mission, etc need to be reviewed, maybe rebuilt to be even safer.
