Teaching a Plane to Dance: Music-Guided Aerobatic Flight (Research Report)

The poster asked whether a machine could make an aircraft dance. This report tells the full story: how a song becomes a flight, how a neural network learns to fly it on the beat, and what the music taught us about physics.

Share this post

Choose a social network to share with, or copy the URL to share elsewhere

This is a representation of how your post may appear on social media. The actual post will vary between social networks

Preview

So, can an airplane dance? In simulation, yes. The system picks out the strongest beats of a song, turns them into a smooth three-dimensional choreography, and a reinforcement-learning pilot learns to fly it, arriving on each beat seven times more accurately than a classical controller.

Its secret is anticipation. Like a dancer who knows what comes next, the pilot sees the upcoming movements and starts each turn early instead of catching up after the beat. Behind every move, a physics model of the aircraft checks that the flight stays within what a real aeroplane can do.

One surprising lesson came from the music itself. How sharp a turn can be depends not only on the strength of the beat and the agility of the aircraft, but also on the time between two turns. A choreography that turns on every beat leaves no time for sharp corners: like a good dancer, a good routine marks the accents rather than every count.

A huge thank you to Ruihan Jia, Hongze Wang and Prof. Dario Floreano for their guidance, and to the Laidlaw Foundation for making this internship possible!