Enhancing Aerial Navigation of Drones Through Imitation of Flying Squirrels' Methods
In a groundbreaking development, South Korean researchers have studied the integration of flying squirrel technology into quadcopters to enhance drone agility and performance in obstacle courses. This innovative approach leverages the natural gliding abilities of flying squirrels and combines it with machine learning.
Flying squirrels, belonging to the tribe Pteromyini, possess large skin flaps (patagium) between their wrists and ankles that aid their gliding from tree to tree. These fluffy creatures are also able to control their flight by using their tail and patagium for air braking, preventing high-speed collisions with tree trunks during their 90-meter flights.
The researchers aimed to replicate this principle in a quadcopter by adding a similar membrane mechanism between its rotors. This led to the development of a new controller, dubbed Thrust-Wing Coordination Control (TWCC), which manages the extending of the membranes in coordination with thrust from the brushless motors. Instead of relying on trial-and-error, the researchers trained a Recurrent Neural Network (RNN) that was first pre-trained using simulation data, followed by supervised learning to refine the model.
During experiments with obstacle avoidance on a test-track, the RNN-based controller demonstrated impressive results compared to a regular quadcopter. However, the range of these flying squirrel drones is reduced due to additional weight and drag. Nevertheless, this technology could pave the way for drones capable of perching on surfaces while executing amazing feats of agility in the air.
Key innovations that contribute to the improved agility and obstacle course performance of these flying squirrel drones include foldable silicone wings that emulate the gliding flaps of flying squirrels, the TWCC algorithm for real-time wing deployment, sensor fusion for precise positioning, and the use of a Physics-Assisted Recurrent Neural Network (paRNN) for accurate aerodynamic modeling.
By exploiting the variability in aerodynamics presented by the silicone membrane wings through the paRNN, the drones equipped with this control system demonstrate superior agility and obstacle navigation compared to traditional quadcopters. Through these adaptations, the flying squirrel-inspired quadcopters boast greater agility, superior obstacle avoidance, and more effective navigation in complex environments.
Science and technology are crucial in the development of the flying squirrel-inspired quadcopters. The researchers use a Recurrent Neural Network (RNN), a key technology, to refine the control system for these drones, improving their agility and obstacle navigation.
