MIT Researchers Boost Tiny Flying Robot Speed by 450 Percent Using AI Control System

Researchers at MIT announced in September 2026 that they had increased the flight speed of insect-scale flying robots by 450 percent, using a new AI-driven control system that mimics how insects process movement in real time.

The research, led by Associate Professor Kevin Chen and Professor Jonathan P. How, addressed a long-standing problem with miniature aerial robots. The robots had soft artificial muscles capable of rapid wingbeats, but their controllers required manual tuning and could not handle complex maneuvers efficiently.

The team developed a two-part AI control architecture to solve the problem. The first component is a model-predictive controller that acts as an expert planner, using a dynamic mathematical model to map out complex flight paths, including sharp turns and aggressive body tilts. This planner is highly accurate but too computationally intensive to run during actual flight.

The second component is a deep-learning model trained through imitation learning. The researchers used the expert planner to train this faster model, which can compute control commands in real time during flight. The result is a system that combines the accuracy of the planner with the speed needed for live operation.

In testing, the robot performed 10 consecutive somersaults in 11 seconds while contending with wind disturbances. It also executed a saccade, a rapid pitching maneuver that insects use to stabilize their vision and adjust position.

The acceleration improvement reached 255 percent compared to previous benchmarks. The researchers say the next phase of development will focus on integrating onboard sensors and cameras so the robots can operate without external motion-capture systems.

The team envisions the technology being used in search-and-rescue operations, particularly in earthquake rubble or narrow spaces where larger drones cannot reach.