RoboFlip
MIT tiny flying robot · AI control · 450% speed boostMIT researchers developed a neural-network flight controller for their insect-scale flying robot — boosting speed by 450% and enabling 10 somersaults in 11 seconds. Toggle between legacy PID and AI control, then try the challenge.
Speed1.0×
Controller
PID
Speed Multiplier
1.0×
Flips Completed
0
Challenge
—
How it works: MIT's robot weighs under 100 mg and flaps four piezoelectric wings at ~120 Hz. The new controller is a three-layer MLP trained entirely in simulation, then deployed onboard. Compared to the legacy fixed-gain PID, the neural net responds faster to disturbances and can plan aggressive rotational maneuvers — trimming 10 consecutive somersaults from ~49 s down to just 11 s.
Control pipeline