Interactive avionics simulators running AITI's actual embedded rendering engine, ported from our STM32 production codebase to your browser.
The panel above is not a mock-up of a product roadmap. It is the same renderer, the same texture and the same geometry that run on the hardware — a 2.761 in round 480 × 480 LCD on an AITI carrier board, with the processor driving the display directly. No GPU and no external graphics part: the sphere, the bank pointers, the flight-director bars, the warning flags, the slip ball and the rate-of-turn pointer are all drawn by the processor, every frame.
The frame figures were measured through the two maneuvers on the panel — a two-turn aileron roll at 60 °/s and a vertical loop — not a static screen. Roll is the worst case rather than the loop, because roll turns the sampling direction against the scanline and pitch does not.
That is the panel on its ribbon, driven by the board behind it, running the firmware. Nothing is pre-rendered and nothing is played back — it is running the same two maneuvers as the buttons on the panel, at the same rates.
One caveat, so it does not mislead you. This is a phone camera under room lighting, and the white balance has pulled the sky toward cyan and flattened the brown. Judge the color from the live face above, not from the video — and treat even that as provisional until a colorimeter has been on a real instrument.
Most glass ADIs cheat: a flat sky/ground image, rotated by bank and slid by pitch. That holds up until the ball carries markings that only make sense on a sphere. A flat image has no third axis for heading to act on and no poles for the meridians to converge at. So this renders an actual textured sphere — every destination pixel traced back through the attitude matrix to a point on the ball.
Done naively that costs 167 ms a frame. Solving the exact spherical case only every eighth pixel and interpolating between anchors, bounding each row by the circle instead of its bounding box, replacing the transcendentals with tables built at boot, and turning the processor's caches back on took it to 27–30 ms — while growing the ball and its texture. That is the difference between a demo and something that goes in a panel.
Not everything on the face has to be drawn, and on a 2.761 in panel not everything can be. The roll-scale ticks, the speed dot column and the index marks are printed on the bezel, not rendered. Turn Bezel artwork off on the panel and you are looking at exactly what the processor draws.
The speed dot column is the clearest case: there is no glass left for it, so the dots are printed and only the white pointer is drawn, moving against them and clear of the ball.
Altimeter, VSI and HSI already ship on this same board — a four-instrument panel from one processor. Tell us the instrument, the interface and the artwork, and we build it to your specification. Talk to us →
Thousands is 1 or blank and units is 0 or blank — the instrument has no other states for those two positions, so the readout steps in tens.
Mutually exclusive by priority bE, UC, AC, OC, FC — the host models that, not the instrument.
;
The serial message that produces exactly what is on the glass, updating as you drive it. Segment counts use : for ten.
These are photographs of a shipped AI1050-2 — one unit, one piece of glass, photographed in both of its presentations. Nothing was swapped between the two shots but the voltage on a lighting pin.
An NVIS (KG3) infrared-absorbing filter sits over the display so the instrument can be viewed through night-vision goggles without blooming, and the day-mode amber is compensated for that filter. The panel above is drawn from these photographs, so it is a fair likeness rather than a dimensional one.
The Smith Industries 20-246-01 uses a segmented liquid-crystal display, custom-etched for that one instrument. When the panel goes, there is no replacement to buy — the tooling is long gone and the quantities were never large enough to justify remaking it. The instrument becomes unsupportable while the trainer around it still has years of service left.
The way out is to stop treating the glass as a fixed part. A round color TFT behind a laser-cut mask can draw the original presentation exactly, and drawing it in software means the same hardware can also draw it in NVIS green for goggle-equipped training — something an etched panel physically cannot do.
Same 2 inch envelope, same rear receptacle in the same location and clocking, same pin assignment, same serial protocol, same self-test behavior. Cabling made for the AI1050-1 it supersedes is used without change, and the operator's existing host software and training procedures stay valid.
Two supply lines the original needed are simply not used any more, and the panel-lighting inputs that used to drive incandescent lamps at up to 2.5 A are now sense inputs drawing under a milliamp — they set the color mode and the brightness instead of lighting anything.
If you are maintaining a trainer with instruments you can no longer source, this is the path. Talk to us →