In the future, small flying robots could be used to help locate survivors trapped under rubble after a major earthquake. Much like insects, these robots would be able to navigate through narrow gaps that larger machines cannot, while avoiding obstacles and falling debris.
Until now, aerial microrobots have only been able to fly slowly along smooth paths, far from the quick and nimble flight of real insects.
However, MIT researchers have now showcased aerial microrobots capable of achieving speed and agility comparable to their biological counterparts. The team developed a new AI-based controller that allows the robotic insect to perform acrobatic maneuvers, such as continuous body flips.
By using a two-part control system that balances high performance with computational efficiency, the robot’s speed and acceleration increased by approximately 450% and 250%, respectively, compared to previous best demonstrations.
Nimble Robot Mimics Insect Moves in Wind
The agile robot was able to complete 10 consecutive somersaults in just 11 seconds, even with wind disturbances threatening to disrupt its flight path.
“We aim to deploy these robots in environments where traditional quadcopters would struggle, but insects could easily navigate. With our bioinspired control system, the robot’s flight performance— in terms of speed, acceleration, and pitching angle—now closely matches that of real insects. This marks a significant step forward toward achieving that goal,” says Kevin Chen, associate professor in the Department of Electrical Engineering and Computer Science (EECS), head of the Soft and Micro Robotics Laboratory at the Research Laboratory of Electronics (RLE), and co-senior author of a paper on the robot.
Chen co-authors the paper with co-lead authors Yi-Hsuan Hsiao, an EECS graduate student at MIT; Andrea Tagliabue, Ph.D.; and Owen Matteson, a graduate student in the Department of Aeronautics and Astronautics (AeroAstro). Other contributors include EECS graduate student Suhan Kim, Tong Zhao, and co-senior author Jonathan P. How, the Ford Professor of Engineering in the Department of Aeronautics and Astronautics and a principal investigator at the Laboratory for Information and Decision Systems (LIDS).
An Artificial Intelligence Controller
Chen’s team has been developing robotic insects for over five years.
Recently, they introduced a more durable version of their tiny robot—a microcassette-sized device that weighs less than a paperclip. This new model features larger, flapping wings, allowing for more agile movements. The wings are powered by a set of soft artificial muscles that flap at incredibly high speeds.
However, the robot’s controller—the “brain” that determines its position and directs its flight—was previously hand-tuned by humans, which limited its performance.
To enable the robot to fly quickly and dynamically like a real insect, it required a more advanced controller capable of handling uncertainty and performing complex optimizations rapidly.
Such a controller would be too computationally demanding to run in real-time, especially given the complex aerodynamics of the lightweight robot.
To address this, Chen’s team partnered with How’s group to develop a two-step, AI-powered control system that combines the necessary robustness for quick, complex maneuvers with the computational efficiency required for real-time operation.
“The hardware improvements enabled us to enhance the controller, allowing for more advanced software capabilities, and as the controller evolved, it opened up more possibilities for the hardware as well. As Kevin’s team pushes the robot’s capabilities, we show that we can leverage them,” says How.
Model-Predictive Controller Enables Complex, Safe Maneuvers
In the first step, the team developed a model-predictive controller. This advanced controller uses a dynamic mathematical model to predict the robot’s behavior and plan the optimal series of actions to follow a trajectory safely.
Although computationally demanding, it can plan complex maneuvers like aerial somersaults, sharp turns, and aggressive body tilting. The high-performance planner also accounts for the forces and torques the robot can apply, which is crucial for preventing collisions.
For example, to perform multiple flips in succession, the robot must decelerate precisely to ensure the initial conditions are perfect for the next flip.
“If small errors accumulate, and you try to repeat the flip multiple times, the robot will just crash. We need robust flight control,” How explains.
They use this expert planner to train a “policy” based on a deep-learning model, which allows the robot to be controlled in real time through a process called imitation learning. The policy acts as the robot’s decision-making system, guiding it on where and how to fly.
In essence, imitation learning converts the powerful controller into a computationally efficient AI model that can operate quickly.
The key challenge was developing an effective method to generate just enough training data to teach the policy everything it needs for executing aggressive maneuvers.
“The robust training method is the key to this technique,” How says.
The AI-powered policy takes the robot’s position as input and generates control commands in real time, such as thrust forces and torques.
Insect-Inspired Performance
In their experiments, this two-step approach allowed the insect-scale robot to fly 447% faster and achieve a 255% increase in acceleration. The robot successfully completed 10 somersaults in 11 seconds, and its trajectory stayed within just 4 or 5 centimeters of the planned path.
“This work shows that soft and microrobots, which previously moved slowly, can now leverage advanced control algorithms to achieve agility comparable to natural insects and larger robots, opening new possibilities for multimodal locomotion,” says Hsiao.
The researchers also demonstrated saccade movement, a behavior where insects rapidly pitch one way to move quickly to a target, then pitch the opposite direction to stop. This swift acceleration and deceleration help insects localize themselves and improve their vision.
“This bio-inspired flight behavior could be crucial in the future as we begin equipping the robot with cameras and sensors,” says Chen.
Equipping the microrobots with sensors and cameras to enable outdoor flight, without relying on a complex motion capture system, will be a key area of future research.
Future Focus
The team also plans to explore how onboard sensors can help the robots avoid collisions with each other and coordinate their movements.
“For the micro-robotics field, I hope this paper marks a paradigm shift by demonstrating that it’s possible to develop a new control system that is both high-performing and efficient,” Chen adds.
“This work is particularly impressive because the robots still manage to perform precise flips and rapid turns, despite significant uncertainties from factors like fabrication tolerances in small-scale manufacturing, wind gusts over 1 meter per second, and even the power tether wrapping around the robot as it executes repeated flips,” says Sarah Bergbreiter, a professor of mechanical engineering at Carnegie Mellon University, who was not involved in the study.
“While the controller currently runs on an external computer rather than onboard the robot, the authors demonstrate that similar, though less precise, control policies can still operate using the limited computational resources of an insect-scale robot.” This is exciting because it suggests that future insect-scale robots could have agility close to that of their biological counterparts,” she adds.
Read the original article on: Tech Xplore