MIT's Insect-Scale Flying Robot Pulls Off 10 Flips in 11 Seconds with AI Control

MIT researchers have pushed insect-scale aerial robotics into a more agile regime: a tiny flapping-wing robot controlled by a deep-learned system performed rapid turns and 10 consecutive flips in 11 seconds. MIT reports the robot moved about 450% faster than the group's earlier system.

The control method combines deep learning with robust tube model predictive control. In plain terms, the learned controller helps the robot choose fast movements, while the robust control layer accounts for uncertainty and keeps the manoeuvre within safe bounds. That matters at this scale: a tiny flying machine has little room for heavy sensors or large onboard computers, and its motion is easily disturbed.

Why do acrobatics matter? Not because a rescue robot needs to do backflips, but because tight turns and rapid recovery are signs of better control in cluttered, unpredictable spaces. The researchers point to future applications such as searching earthquake rubble or navigating other hazardous environments where conventional drones may struggle.

The research is reported by MIT and summarized by ScienceDaily. The underlying paper, by Yi-Hsuan Hsiao and colleagues, was published in Science Advances in 2025; the September 2026 MIT news item highlights the system's capabilities. So this is a newly surfaced research story, not a paper published this week.

The next hurdles are practical: carrying useful sensors or payloads, sustaining flight time, handling real-world airflow and debris, and proving reliable autonomy outside controlled tests. A flip video is a compelling demo; dependable search-and-rescue performance is a much higher bar.