ChatGPT as ATC

  • Just wanted to share something that provides an ATC experience for those who are looking for that. I have installed ChatGPT on my Mac (but should be the same for PC) and use it as ATC client for Aerofly. Of course it’s not perfect; it doesn’t know your location so isn’t very interactive and ChatGPT has to learn from you, so the first few times I was correcting it a bit and sometimes still have to. But I have to say, it’s a really nice and free way of adding some more immersion to the already great sim!

    You can call up the pop up in game with a shortcut on your keyboard, almost like if it is fully integrated in game. I let it create a flight plan or tell it my flightplan, it even knows that Aerofly requires manual weather set up and advises on that. Then it actually works like a charm, though you always have to trigger a next message, for example when passing a certain waypoint. It even works with speech recognition, quite easy to set up.

    Just wanted to share it so others can try it as well. Give it a shot! Also for flight plan creation it is pretty awesome, just ask it for a challenging flight and it will prove you with interesting options.


    Sometimes it just is a bit too friendly, or lacking any trust in my flying skills though; ATC wishes me good luck on landing the 737 in Madeira. ^^

    Edited 3 times, last by Mtyn (January 4, 2025 at 3:43 PM).

  • Looks like it does, cool! I trained it the first few times. You have to turn ChatGPT memory on, so it remembers what you tell it. Then ask it to simulate ATC as a tower controller for you during this flight, on a flight from x to x, with the route you plan to fly.

    I usually ask it for a nice flight / route, then the waypoints (I like GA, so let it search for VOR and NDB routes) and then I just ask it to simulate ATC for me during the flight. And on you go!


    Everytime it says something strange or you want it to do something different just tell it how you want it, and next time it is better. The one thing I found is that the courses between VOR’s are totally off, so I now ask it to only give me courses if it’s 100% sure. But it should be pretty fast adapted, took me 2-3 flights with some remarks. You can even learn it how Aerofly works (manual conditions) and it will advise how to set those up. Pretty cool!

  • You know what? This is the kind of setup I'm using. My Xiaomi 13T running AFG, my older phone running flight sim tools I created (I could share the link but it's something related to promo and idk if it's allowed here), my iPad running checklists and nav and my PC running ChatGPT.

    Aviationly,

    SkyX

  • You know what? This is the kind of setup I'm using. My Xiaomi 13T running AFG, my older phone running flight sim tools I created (I could share the link but it's something related to promo and idk if it's allowed here), my iPad running checklists and nav and my PC running ChatGPT.

    Nice! The good thing about it and why I wanted to share it, is that the ChatGPT app can pop-up. So especially for users running on one device this works really well, almost like it is integrated in Aerofly.

    But yeah, no matter where you run it ChatGPT is pretty capable of acting as a good ATC client!

  • Great idea! Take a look at this, it may be interesting to you:

    https://airplane.team/

    I am subscribed, if you have any questions.

    That looks really nice as well, though it seems that with ChatGPT you can get most of this for free as there is no Aerofly integration as far as I see it? Would be amazing if it could use Aerofly’s location data it sends on your network. I think an AI ATC sim app would be all we ever need, not requiring an ATC development in Aerofly itself (unless it also communicates with the AI planes flying around).

  • That looks really nice as well, though it seems that with ChatGPT you can get most of this for free as there is no Aerofly integration as far as I see it? Would be amazing if it could use Aerofly’s location data it sends on your network. I think an AI ATC sim app would be all we ever need, not requiring an ATC development in Aerofly itself (unless it also communicates with the AI planes flying around).

    I will ask the nice team of airplane.team if they can do a basic integration with Aerofly FS4. Actually they are doing the integration with MSFS2024, and they started the product with X-Plane. They have an active Discord channel where we discuss about the product. Let's hope the best!

    Best regards, Juan

  • I tried it, but it's more of a gimmick than an ATC. It's good for quick information (frequencies, etc.). I hope that IPACS installs something really useful at some point.

    It sure isn’t a full ATC experience, but I have to say that it actually does a pretty decent job if you ask it to simulate an ATC role (see the screenshots). Giving you alle the clearances, altitudes during flight, STAR’s and SID’s. It just doesn’t know your position so you have to tell it you pass a certain waypoint so it sort of knows where you are for the upcoming instructions.

    But yeah, sure don’t want to present this as an ATC but it does provide some more immersion. Maybe an AI model is easier to integrate sometime then a completely programmed ATC, specially if it would know your location and the location of other airplanes.

    For example, this is my recent text dump of the chat I had for a flight from Schiphol to Billund including how i start it. I liked this one as it even corrected me if I was wrong (using speech recognition with a Dutch accent isn't ChatGPT's favorite):

    Martijn: Hi! I'm going for a sim flight from EHAM (27) via BERG1P SID to EKBI 27 where I will expect an ILS arrival. I use the 737-900ER, FL240, IFR. Wind 235/15. Can you do the ATC simulation?
    ChatGPT: Sure! You can start with the ground frequency for taxi-instructions. Let me know when you're ready!

    Martijn: Amsterdam Delivery, KLM025 is requesting IFR clearance to Billund.
    ChatGPT: KLM025, Amsterdam Delivery, good afternoon. Clearance on request, standby.
    *Moment.. clearance is defined...*
    KLM025, cleared to Billund via the BERG1P departure, climb to FL240, squawk 4325. Departure frequency 121.200.
            
    Martijn: Amsterdam Delivery KLM 025, clear to Billund via de BERG1P departure, climb to flight level 240, squawk 4325.
    ChatGPT: KLM025, readback correct. Contact Ground on 121.805 when ready for pushback and start.

    Martijn: Amsterdam Ground KLM025 ready for pushback and start.
    ChatGPT: KLM025, Amsterdam Ground, pushback and start approved, face west.

    Martijn: Pushback and start approved, facing west, KLM 025.
    ChatGPT: KLM025, report ready for taxi.

    Martijn: KLM 025 ready for taxi. Runway 27.
    ChatGPT: KLM025, taxi to runway 27 via Sierra 4, Bravo, and Alpha. Hold short of runway 27.

    Martijn: Taxi to runway 27 via CR4, B and A, holding short of runway 27, KLM 025.
    ChatGPT: KLM025, correction: taxi via Sierra 4 (S4), Bravo (B), and Alpha (A). Hold short of runway 27.

    Martijn: Taxi to runway 27 via Sierra 4 , Bravo and Alfa, holding short of runway 27, KLM 025.
    ChatGPT: KLM025, readback correct. Contact Tower on 119.225 when holding short of runway 27.

    Martijn: Amsterdam Tower, this is KLM 025, holding short runway 27, ready for departure.
    ChatGPT: KLM025, Amsterdam Tower, good afternoon. Hold short of runway 27, traffic landing.

    Martijn: KLM 025, hold short for runway 27.
    ChatGPT: KLM025, traffic landed, line up and wait runway 27.

    Martijn: Line up and wait for runway 27, KLM 025.
    ChatGPT: KLM025, cleared for takeoff runway 27, wind 235 at 15 knots.

    Martijn: Cleared for takeoff runway 27, KLM 025.
    ChatGPT: KLM025, you are now airborne. Contact Departure on 121.200, have a good flight.

    Martijn: Contact departure on 121.2 Goodday! KLM 025.
    ChatGPT: KLM025, good day!

    Martijn: Amsterdam Departure KLM 025 climbing through 2,000 feet for flight level 240 via the BERG1P departure.
    ChatGPT: KLM025, Amsterdam Departure, roger. Continue climb FL240, proceed direct BERGI, radar contact.

    Martijn: Continue flight level 240, direct BERGI. KLM025.
    ChatGPT: KLM025, continue direct BERGI and climb to FL240. Squawk 4325, contact Bremen Control on 127.375 for enroute.

    Martijn: Bremen Control KLM 025 is with you at FL240.
    ChatGPT: KLM025, Bremen Control, good day. Radar contact at FL240. Continue on present heading and report when reaching the EKBI ILS approach.

    Martijn: Continue present heading and reporting when reaching the Billund ILS. KLM025.
    ChatGPT: KLM025, roger. I’ll be here when you reach the Billund ILS.

    Martijn: Bremen Control requesting descent for approach KLM 025.
    ChatGPT: KLM025, Bremen Control, descend to FL150 for the EKBI ILS approach.

    Martijn: Flight level 1-5-0, KLM 025.
    ChatGPT: KLM025, descend to 5,000 feet, expect ILS approach runway 27 at Billund.

    Martijn: Descend to 5000ft, expecting ILS approach runway 27 at Billund, KLM 025.
    ChatGPT: KLM025, Billund Approach, good day. You are cleared for the ILS approach runway 27, report when established on the localizer.

    Martijn: Billund Approach, KLM025, cleared for the ILS approach runway 27, report established on localizer.
    ChatGPT: KLM025, roger. ILS approach runway 27 cleared, report established.

    Martijn: Established on the localizer runway 27, KLM025.
    ChatGPT: KLM025, Billund Approach, maintain 5,000 feet until the glideslope intercept, cleared to continue the ILS approach runway 27.

    Martijn: Maintain 5000ft until glidescope intercept. Cleared to continue the ILS approach runway 27, KLM 025.
    ChatGPT: KLM025, Billund Approach, readback correct. Contact Billund Tower at 118.500.

    Martijn: Billund Tower, this is KLM 025 on final runway 27.
    ChatGPT: KLM025, Billund Tower, wind 235 at 15 knots, runway 27 cleared to land.

    Martijn: Wind 235 at 15 knots, runway 27 cleared to land, KLM 025.
    ChatGPT: KLM025, landing confirmed. Have a smooth touchdown, and welcome to Billund!

    Martijn: Billund Ground, KLM025 cleared runway 27. Taxi to gates.
    ChatGPT: KLM025, Billund Ground, taxi to gate via Bravo, hold short of Taxiway Charlie.

    Martijn: Taxi to gate via Bravo, holding short of taxiway Charlie. KLM025.
    ChatGPT: KLM025, readback correct.

    Martijn: Holding short at taxiway Charlie, KLM025.
    ChatGPT: KLM025, continue taxi to gate via Bravo, contact gate operator for parking instructions.

    Martijn: Continue taxi via Bravo. Thank you and good night! KLM025.
    ChatGPT: KLM025, you’re welcome! Have a great night and enjoy your time at Billund!

    Edited 2 times, last by Mtyn (January 6, 2025 at 11:32 PM).

  • Mtyn, if you want to get the location and some other flight params (latitude, longitude, altitude, ground speed, heading, pitch, and roll), please take a look at my map app. The most interesting part it has it is that it read the UDP stream AFS4 can produce, and use it to locate a plane in the map. Maybe you can do something with this, like capturing a snapshot of the UDP data when the prompt is used, and injected it as part of the prompt. Link to the app in my signature, and link to Github code here:

    Python Aircraft Tracker on map for Aerofly FS4 (Copyright IPACS) Flight Simulator
    Python Aircraft Tracker on map for Aerofly FS4 (Copyright IPACS) Flight Simulator - aerofly_fs4_maps_v25.py
    gist.github.com

    Best regards, Juan

  • Glad you like the idea!

    Please refer to this old post to see more information about this app and the UDP stream (specially the Foreflight specs):

    Jugac64
    September 2, 2024 at 7:51 PM

    Best regards, Juan