In a remarkable scientific achievement, MIT researchers have developed aerial microrobots capable of performing complex aerial maneuvers with unprecedented speed and agility. These tiny flying robots, weighing less than a paperclip, can now execute movements comparable to their biological counterparts—real insects—potentially revolutionizing search-and-rescue operations in disaster scenarios.
The Leap Forward in Microrobot Performance
The MIT team, led by Associate Professor Kevin Chen from the Department of Electrical Engineering and Computer Science, has achieved a staggering 447% increase in speed and 255% improvement in acceleration compared to previous iterations of their flying microrobots. This dramatic enhancement enables the tiny machines to perform feats previously thought impossible at this scale, including completing 10 consecutive somersaults in just 11 seconds while maintaining precise trajectory control within 4-5 centimeters.
The breakthrough stems from a novel AI-based control system that combines computational efficiency with high performance, allowing the robots to mimic the swift, agile movements characteristic of real insects. This advancement represents a significant step toward deploying these robots in practical applications where larger drones cannot operate effectively, such as navigating through narrow passages in collapsed buildings.
Innovative Two-Step AI Control System
At the heart of this achievement lies a sophisticated two-part control scheme developed through collaboration between Chen’s Soft and Micro Robotics Laboratory and Professor Jonathan How’s team. The system consists of:
First, a model-predictive controller that uses dynamic mathematical modeling to predict robot behavior and plan optimal movement sequences. This controller accounts for physical constraints and ensures the robot can safely execute complex maneuvers like aerial somersaults and aggressive body tilting.
Second, an imitation learning process that compresses this powerful controller into a computationally efficient AI model. This approach enables real-time decision-making without sacrificing performance—a critical balance for microrobots with limited onboard computing capabilities.
The researchers describe this robust training method as the