Researchers Turn to a Highly-Efficient Deep Learning Model to Deliver a Bot as Agile as a Bee

Researchers from the Massachusetts Institute of Technology (MIT) have designed a tiny aerial robot that, they say, is as fast and agile as a bumblebee — and which could one day help in search-and-rescue missions.

"We want to be able to use these robots in scenarios that more traditional quadcopter robots would have trouble flying into, but that insects could navigate," co-senior author Kevin Chen, associate professor and head of the Soft and Micro Robotics Laboratory in MIT's Reserach Laboratory of Electronics, explains of the project. "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. This is quite an exciting step toward that future goal."

The robot itself is based on earlier work on tiny aerial robots, modified to have larger flapping wings — like those of a bee — driven by soft artificial muscle actuators to provide improved agility. The potential performance of the robot itself, though, outstripped the capabilities of its "brain," forcing the team to develop a two-step artificially intelligent (AI) control system that can handle high-speed maneuvers without needing excessive computational resources.

"The hardware advances pushed the controller so there was more we could do on the software side, but at the same time, as the controller developed, there was more they could do with the hardware," explains co-lead author Jonathan P. How. "As Kevin [Chen]'s team demonstrates new capabilities, we demonstrate that we can utilize them."

The system, a model-predictive system that trained a computationally-efficient deep-learning policy model through imitation learning in order to provide real-time control, proved its capabilities in testing: the tiny robot was able to fly 447 percent faster with a 255 percent increase in acceleration than under the hand-tuned model's control, completing 10 somersaults in 11 seconds and never straying more than a couple of inches off its planned trajectory. The system was also used to implement saccade movement, the rapid flitting used by some insects for localization.

"This bio-mimicking flight behavior could help us in the future when we start putting cameras and sensors on board the robot," Chen says. "For the micro-robotics community, I hope this paper signals a paradigm shift by showing that we can develop a new control architecture that is high-performing and efficient at the same time."

The team's work has been published in the journal Science Advances under open-access terms.