Lisbon in the R66 — with an AI copilot riding along in VR (52 min, no cuts)

Community Content Hangar Videos & Streams
, , , ,
1

I fly with an AI voice copilot now. This is 52 minutes of Lisbon with her.

Robinson R66, MSFS 2024, flown in VR. No time compression, no cuts — one hop, out to Cascais and back over the city. What’s different is the second voice in the cockpit.

What she is

I hold a trigger and ask. “What’s that mural on the building?” — and she tells me, because she can see what I’m looking at. She knows the aircraft, the altitude, the heading, and where the nose is pointed, and she looks things up while we fly. What she names goes on a map on the tablet in the cockpit.

What she isn’t: ATC. No frequencies, no phraseology, no “cleared to land.” She’s company. She rides along and talks about what’s under the skids.

What Lisbon gave her

The city is a good test because it stacks things fast. In one pass down the river you get the Ponte 25 de Abril, Alcântara underneath it, and the whole Baixa opening up behind. She named the Aqueduto das Águas Livres and I sat there remembering a taxi ride under it years ago. Castelo de São Jorge — I’ve actually stood on those walls, and hearing the history read back to me at 500 ft is a strange and very good feeling. Praça do Comércio, the azulejos, the calçada. Murals from the air that you’d never see from the ground.

Before takeoff I had her drop pins for pools, restaurants, parks and pharmacies just to see the map fill up. It filled up.

And the part that didn’t work

She marked the Elevador da Glória for me and I flew straight over it without ever spotting it. Then I told her to cancel the destination — three different ways — and it wouldn’t cancel. I had to shut the whole thing off. It’s in the video with its own chapter, because that’s the deal I made with this series: what works goes up, what breaks goes up too. (For the curious: the marker went to the map but the leg never armed, because the target was 220 m away and the floor is 340 m. Fixed since.)

The VR setup — the question I get most

  • PC: RTX 4090 · i9-13900K · 96 GB RAM
  • Headset: Meta Quest 3
  • Link: Virtual Desktop over USB 3.0 at 5120 Mbps. I moved off Wi-Fi 6 (2400 Mbps) and the difference is not subtle — if you fly wireless and have a spare USB 3 cable in a drawer, it’s worth ten minutes of your evening.
  • Performance: a steady 85 fps over Lisbon photogrammetry, with all of the above running alongside the sim.
  • Recording: NVIDIA GeForce Experience

Ask me anything about any of it.

The series, and the beta

Track On Air is a limited run of test flights recorded while the app gets built. Human and AI in the same cockpit, tuning the conversation as we go. This is S1E16; there’s a playlist on the channel.

→ VRFlights73 — https://www.youtube.com/@VRFlights73/videos

There’s no version you can install at home yet. When the beta opens I’ll have a small number of keys. If you want one, email rvalle7@icloud.com with Track as the subject and I’ll put you on the list — or just say so in this topic and I’ll find you.

Happy to answer anything about how it works under the hood. It’s been the most fun I’ve had building something in a long time.

2

Very cool concept! So I assume this supports just about every language and you can ask detailed questions about landmarks etc? Since it’s just a general LLM in the background? Will it be a subscription or how are the tokens and TTS financed?

3

Thanks — those are the right questions.

On language: the model in the background can talk in a lot of them. The product around it can’t, not yet. Voice commands, speech recognition, the voice itself and the map labels all have to be built and flight-tested per language, so I’m not going to claim “every language.” Portuguese is what’s in the videos because that’s what I fly and debug in. English is next, then Spanish, French and German.But the Heart, STT and TTS are all Grok API, só, she knows a lot of languages already.

On landmarks: that’s the whole point. You hold a trigger, look at something, and ask. She has GPS, the aircraft, heading, altitude, and — this is the part that isn’t a generic chatbot sitting next to the sim — the direction you’re looking in VR. Then she looks the place up (maps, Wikidata, the web) while we fly, and the name drops onto the tablet in the cockpit. Follow-ups work: history, who built it, what’s that mural, is that the same bridge we crossed ten minutes ago. If she isn’t sure, she says so. That’s a rule, not a vibe.

She isn’t ATC. No frequencies, no clearances. There are already good products for that.

On money: Track itself is a one-time purchase, not a subscription. The tokens are yours. You make an xAI API account, put prepaid credit on it, paste the key into Track, and the conversation, the speech-to-text and the voice all run on that. SuperGrok / X Premium do not cover this — different door, different bill. The credit sits there until you fly; it isn’t a monthly plan that resets. Measured over a few dozen real flights, an hour of talking in the cockpit is well under a dollar.

There’s no installer yet. If you want a beta key when there is one, email rvalle7@icloud.com with Track as the subject — or drop your email here and I’ll put you on the list.

4

Unfortunately I don’t have time currently to do beta-type testing and will be on vacation the next few weeks, but I’ll follow this thread so I don’t forget it once released fully.

5

Don´t worry. I have a site where you can put your email and Name to enter the waiting list.
That way you will be email when the final release is set.
track73.com.br
Chears!

6

Nice, joined the list.