Indiana-based Purdue University said on 9 May that researchers had engineered drones that “behave like hummingbirds”, trained by machine learning algorithms based on various techniques the bird naturally uses every day.
This means that – after learning from a simulation – the robot “knows” how to fly and hover on its own like a hummingbird would, such as discerning when to perform an escape maneuver, allowing them to move “better through collapsed buildings and other cluttered spaces to find trapped victims”.
A combination of AI and flexible flapping wings allows the robot to “teach itself” new tricks. For example, it can’t yet “see” but senses by touching surfaces. Each touch alters an electrical current, which the researchers realized they could track.
“The robot can essentially create a map without seeing its surroundings,” Xinyan Deng, an associate professor of mechanical engineering at Purdue, said in a statement. “This could be helpful in a situation when the robot might be searching for victims in a dark place – and it means one less sensor to add when we do give the robot the ability to see.”
While drones cannot be made infinitely smaller as they would be unable to generate enough lift to support their weight, hummingbirds do not use conventional dynamics and have extremely resilient wings.
“The physics is simply different; the aerodynamics is inherently unsteady, with high angles of attack and high lift,” Deng said. “This makes it possible for smaller, flying animals to exist, and also possible for us to scale down flapping wing robots.”
According to Purdue, researchers have long tried to “decode” hummingbird flight to facilitate robots flying where larger aircraft cannot. For example, California-based drone developer and manufacturer AeroVironment was commissioned by DARPA – a US Department of Defence agency – in 2011 to build a robotic hummingbird that was heavier than a real one but not as fast. It had helicopter-like flight controls and limited manoeuvrability, and required a human to operate the remote control at all times.
Deng’s research group and fellow collaborators studied hummingbirds themselves over multiple summers in Montana, documenting key hummingbird manoeuvres, such as rapid 180-degree turns, and translated them into computer algorithms that a computer could learn from when hooked up to a simulation.
Further study of the physics of insects and hummingbirds allowed researchers at Purdue to create robots that are smaller than hummingbirds – and even as small as insects – without compromising the way they fly. According to Deng, a smaller drone with a greater wing flapping frequency will fly most efficiently.
The drones have 3D-printed bodies, wings made from carbon fibre and laser-cut membranes. One weighs as little as 12 grams – the weight of the average adult hummingbird – and another insect-sized drone weighing just one gram. The hummingbird-sized drone can lift up to 27 grams.
Drones with higher lift give researchers more room to eventually add a battery and sensing technology, such as a camera or GPS, and while the drone currently needs to be tethered to an energy source, the researchers reportedly say that this will not be the case for much longer. It only requires two motors to fly and can independently control each wing, which is how flying animals are capable of performing highly agile manoeuvres in nature.
The drones can fly as silently as a real hummingbird, making them ideal for covert operations, and can stay steady through turbulence, which researchers demonstrated by testing the dynamically scaled wings in an oil tank.
Robotic hummingbirds could both help with search-and-rescue missions and allow biologists to study hummingbirds more reliably in their natural environment through the senses of a realistic robot.
“An actual hummingbird has multiple groups of muscles to do power and steering strokes, but a robot should be as light as possible, so that you have maximum performance on minimal weight,” Deng said. “We learned from biology to build the robot, and now biological discoveries can happen with extra help from robots.”
Early stages of the work, including the Montana-based experiments in collaboration with researchers from the University of Montana, were financially supported by the National Science Foundation. The project is related to Purdue’s 150th anniversary Giant Leaps celebration, acknowledging the university’s global advancements in AI, algorithms and automation as part of Purdue’s 150th anniversary.
The researchers will present their work on 20 May at the 2019 IEEE International Conference on Robotics and Automation in Montreal with a YouTube video available here if you’re unable to attend. Simulations of the technology are available open-source on GitHub.
