CASE_STUDYAVIATION

The Architecture of Autonomous Flight

How we built a neural-symbolic hybrid system to control manned aircraft in real-time.

"The neural-symbolic hybrid is the answer to a tension as old as aviation itself."
It learns what it can’t be programmed for, and it’s constrained by what must never be violated.

Client
Confidential
Date
January 2026
Industry
Aviation

Challenge

Traditional autopilots rely on rigid state machines. They work well when conditions are predictable, but fail catastrophically in edge cases — and in aviation, catastrophic failure has a very literal meaning.

Outcome

A hybrid neural-symbolic system running at 100Hz that can see and react like a human pilot while following safety procedures with machine precision. Demonstrated autonomous takeoffs with zero human input.

Technologies

RustC++PythonROS 2PX4YOLOv8ACAS-XuPPODreamerV3NAACSOFIA

The Problem with Autopilots

Traditional autopilots are state machines. They follow rigid decision trees: if altitude drops below X, pitch up by Y degrees. This works when the world behaves as expected. It falls apart when it doesn't.

Edge cases kill. A sudden wind shear, an unexpected obstacle, a sensor reading that doesn't fit the programmed model — these are the moments where rigid systems fail catastrophically.

We needed a system that could handle the unpredictable. One that could see the world, interpret it, and react — all within milliseconds.

100Hz
Control Loop
ACAS-Xu
Safety Rules
YOLOv8
Vision Model
ZERO
Human Input

Technical Breakdown

The following deep-dive covers the full architecture — how the dual neural-symbolic system works at 100Hz, the control loop in Rust, and the live flight test results.

Blog post "001-architecture-autonomous-flight" not found

With the core architecture established, the next question was: how do you train an AI to actually fly? Not through programming, but through practice.

Teaching AI to Fly

Our reinforcement learning approach mirrors how human pilots learn — through trial, error, and progressively harder scenarios. The AI that practiced in storms became the AI that handled anything.

Blog post "learning-to-fly" not found