Researchers at the Massachusetts Institute of Technology (MIT) have developed a micro-flapping-wing robot that exhibits flight characteristics and speeds similar to those of a bee. An AI-based, bio-inspired flight control system enables the flapping-wing robot to fly along consistent trajectories and perform acrobatic maneuvers such as somersaults.

The MIT flying robot measures only 40 mm x 40 mm x 9 mm and weighs 750 mg, less than a paperclip, the MIT scientists write in the study „Aerobatic maneuvers in insect-scale flapping-wing aerial robots via deep-learned robust tube model predictive control“, which was published in Science Advances. The robot is propelled into the air by four independently operating flapping wings, each moved by a dielectric elastomer actuator (DEA) at a flapping frequency of 330 Hz. Only the deflections of individual wings are altered to control direction. The DEA essentially consists of dielectric elastomer layers combined with thin plastic nanotube electrodes. The robot’s position in space is monitored via an external motion capture system, meaning the flying robot does not require onboard sensors, but is currently only able to fly in the laboratory.

Flight Controller with Artificial Intelligence

To emulate insect flight capabilities, high speeds, braking maneuvers, and rapid changes in direction are necessary. To achieve this, an efficient and fast-acting flight controller had to be created.

Initially, the researchers developed a model predictive controller that uses a dynamic mathematical model to predict the robot’s behavior and plan the optimal sequence of actions, allowing it to safely follow a flight path. The planner takes into account hardware system limitations, such as the force and torque the robot can generate.

Based on this planner, the researchers trained a policy based on a deep learning model. This policy is used for the robot’s decision-making, telling it where and how to fly. With it, they controlled the imitation learning process to create a computationally efficient AI model for a high-performance flight controller.

Compared to the previous model of the micro-flapping-wing robot, this new approach allows the robot to fly 447 percent faster and accelerate 225 percent faster. The robot thus reaches a speed of up to 7.1 km/h and accelerates at a maximum of 11.7 m/s². This enables it to perform demanding acrobatic flight maneuvers, including aerial leaps, rapid turns, and somersaults.

For example, the flapping-wing robot performed ten somersaults within 11 seconds. During these maneuvers, the robot deviated from its flight path by only four to five centimeters. The researchers were also able to make the robot perform sudden acceleration and braking maneuvers. Insects use these to orient themselves and perceive their environment better.

The MIT scientists believe that this could also help the robot orient itself in space once cameras and other sensors are integrated directly into the robot. Then, an external motion capture system would no longer be necessary, and the micro-flapping-wing robot could fly freely outside the laboratory, for example, to be deployed in rescue missions.

(olb)

Don’t miss any news – follow us on Facebook, LinkedIn or Mastodon.

This article was originally published in German. It was translated with technical assistance and editorially reviewed before publication.