Researchers from Purdue University have been working on drones powered by AI that can mimic some of the abilities that hummingbirds possess. The team believes if they were able to engineer a drone with the abilities of a hummingbird, like the ability to hover and change direction quickly, the drones could be helpful in disaster situations. The drones were trained using machine learning algorithms to use techniques that the tiny bird uses naturally every day.
Researchers say that by learning from simulation the robot “knows” how to move on its own as the hummingbird would including the ability to perform an escape maneuver. The drone aircraft uses flexible, flapping wings and its AI allows the robot to learn new skills.
The robot can’t see, but it can sense when it touches a surface, something that the researchers can track easily. The team notes that the robot can “essentially create a map” without having to see its surroundings. This is seen as a potential benefit when the robot might be searching in a dark places for victims. The ability to see by touching surfaces also eliminates the need to add another sensor to allow the robot to see its surroundings. Fewer sensors means less weight, less power consumption, and less cost.
One benefit of basing the drones on hummingbirds is that the bird doesn’t use conventional aerodynamics to fly. Using the same unconventional examples of aerodynamics allows the robot to scale down flapping-wing robots. Some of the team of researchers have spent multiple summers studying hummingbirds in Montana.
They used that study to translate key hummingbird maneuvers into computer algorithms the robot could learn when connected to a simulation. Study of the physics of hummingbirds allowed the researchers to build robots smaller than hummingbirds and as small as insects; research into this type of drone is ongoing.