Sector 03 / Intersection
Aviation is the hardest possible environment for machine learning: safety-critical, heavily regulated and unforgiving of unexplained behaviour. That is exactly what makes this intersection interesting.
7 curated links
Certified autonomy is arriving through cargo and retrofit programmes first. The regulatory question is not whether the software can fly the aeroplane, but how you demonstrate a learned system behaves predictably across the full flight envelope.
Graph neural network forecasts now match or beat traditional numerical models at a fraction of the compute, which changes dispatch, fuel planning and turbulence avoidance.
Sensor streams from engines and airframes feed models that flag component degradation before it becomes an AOG event — the least glamorous and most commercially valuable use of AI in aviation.
Trajectory prediction, conflict detection and arrival sequencing are all being augmented by learned models under EUROCONTROL and FAA programmes.
Autonomy, AI-assisted air traffic and next-gen airframes.
How machine learning is entering European air traffic management.
Certification thinking for machine learning in safety-critical avionics.
Automated flight systems for existing cargo aircraft.
Electric air-taxi aircraft development, certification progress and flight testing.
Autonomous pilot systems targeting certified operations.
ML weather models now beating numerical forecasts — directly useful in dispatch.
Each link opens the actual paper, report or announcement.
DeepMind's model that outperforms traditional forecasts used in flight planning.
A second landmark AI weather model, published with full validation against ECMWF.
Europe's guidance on how machine-learning applications can be certified for flight.
The FAA's published plan for approving learned systems in aviation products.
How artificial intelligence is being introduced into European air traffic management.
Vision-based autonomous taxi, take-off and landing demonstrated on an A350.
Peer-reviewed confirmation that graph neural networks beat operational forecasting.
