AI Control Lets Microbot Match Bumblebee Agility
That’s according to Jonathan P. How, describing the breakthrough that has pushed insect-scale aerial robots into a new performance class. For years, engineers have tried to replicate the blistering speed and nimbleness of flying insects, but microrobots have lagged far behind-until a recent collaboration at MIT delivered a leap in both hardware and control intelligence.
The team’s most recent project is a microcassette?sized flapping?wing robot weighing less than a paperclip. Powered by soft artificial muscles that are able to drive larger wings at high rates, the latest design seemed to have the potential for insect?like maneuvers. But its earlier hand?tuned controller constrained performance. So the challenge was clear: Come up with a control architecture that could manage the robot’s complex aerodynamics, large model uncertainties, and disturbances, while still remaining computationally efficient enough for real?time use.
The solution was a two?step AI?driven control scheme: In the first stage, a high?fidelity model?predictive controller (MPC) plans aggressive trajectories somersaults, rapid turns, and sharp pitch changes while respecting force and torque limits to avoid collisions. This MPC predicts the robot’s dynamic behavior while optimizing each maneuver, including the precise deceleration required to repeat flips without accumulating error. In the second stage, the team compresses this computationally heavy “expert” planner into a lightweight neural network policy through imitation learning. This policy takes the position and attitude of the robot as inputs and outputs thrust and torque commands at high speed, thus enabling the execution of complex aerobatics without the burden of solving intensive optimizations mid?flight.
The ensuing performance gains are astounding: flight speed increased by 447% and acceleration by 255% with respect to previous demonstrations. The flying robot completed 10 consecutive somersaults in 11 s, deviating only 4–5 cm from its intended path, while encountering wind gusts of over 1 m/s. It executes saccade maneuvers, meaning rapid pitch-and-stop movements employed by insects to localize and reduce motion blur, similar to what biological organisms do to navigate in cluttered environments.
This control architecture is in line with advances in robust tube-based MPC for autonomous navigation, whereby predictive models and adaptive feedback keep aerial systems safe in dynamic obstacle-rich settings. By training the neural network within a disturbance-invariant “tube” around the nominal trajectory, the team from MIT ensured resilience against fabrication tolerances, aerodynamic uncertainties, and even tether entanglement during flips.
Hardware innovations enabling this agility draw from the field of bioinspired flight control. Its soft, dielectric elastomer actuators flap at 330 Hz, faster than many insects, for rapid modulation in thrust within millisecond timescales. This responsiveness is combined with the precision of the AI controller to enable acceleration and body rotation magnitudes comparable to those of fruit flies, long considered benchmarks of aerial agility.
Future work targets autonomy outside of the motion?capture lab. The addition of micro?cameras, optical flow sensors, and time?of?flight modules has the potential to empower untethered navigation outdoors, with onboard computation that handles obstacle avoidance, multi?robot coordination, and more. Lessons learned from autonomous navigation in constrained indoor environments point towards further integrating heuristic path planning with optimal tracking to ensure safety in disaster?response scenarios where robots may have to navigate through collapsing structures or debris fields.
To the micro?robotics community, this achievement is a paradigm shift: high?performance and efficiency are not necessarily mutually exclusive. As Kevin Chen put it, “Now, with our bioinspired control framework, the flight performance of our robot is comparable to insects in terms of speed, acceleration, and the pitching angle.” With continued refinement, such systems could bring insect?scale UAVs into real?world missions in search of survivors in rubble, pollination of crops, or exploration of hazardous spaces while matching the agility of their natural flyers.
