MIT Engineers Achieve Bumblebee-Level Speed in Aerial Microrobots

Researchers at the Massachusetts Institute of Technology (MIT) announced a significant development in micro-scale aerial robotics in 2025, demonstrating agile microrobots capable of flight speeds and acrobatic maneuvers comparable to those of insects like the bumblebee. This advance addresses previous limitations in micro-robotics characterized by slow and predictable flight paths, establishing a new standard for performance at this scale.

The core of the innovation lies in a novel, bioinspired control framework integrated with an artificial intelligence-based controller. This system allows the diminutive robotic insect to execute highly dynamic actions, including continuous body flips. The implementation of this advanced control scheme resulted in substantial performance gains: the robot's flight speed increased by approximately 450 percent, and its acceleration rose by about 250 percent compared to earlier iterations.

To quantify the enhanced agility, the device, which is cassette-sized and weighs less than a paperclip, successfully completed ten consecutive somersaults in just eleven seconds while maintaining its intended path despite external wind disturbances. This performance level, which incorporates a saccade—a rapid burst-and-brake action used by insects for sudden stops—was achieved despite the robot's low inertial mass and inherent sensitivity to environmental forces.

The system is propelled by flapping wings driven by artificial muscles, a design focus for Professor Kevin Chen's group at MIT over the past five years. Co-senior author Kevin Chen characterized the achievement as a vital step toward the long-term objective of navigating environments inaccessible to conventional quadcopters. The research, which also involved co-senior author Jonathan P. How from the Department of Aeronautics and Astronautics, was formally published in the journal Science Advances on December 3, 2025.

The AI controller employs a dual architecture: an expert Model Predictive Controller (MPC) for planning aggressive maneuvers within physical limits, and a deep-learned policy trained through imitation learning for efficient real-time execution. This enhanced agility holds potential for time-critical applications, moving aerial microrobotics closer to practical use in confined settings, such as search-and-rescue operations under collapsed rubble where larger drones cannot operate.

Future development, as outlined by the research team, centers on achieving fully autonomous operation by integrating onboard sensors and cameras, moving away from external motion-tracking systems. This transition will be essential for deployment in outdoor environments and for enabling coordinated swarm behavior, signaling a potential shift in the micro-robotics community toward computationally efficient control for insect-scale systems.