Can bio-inspired wind sensing harnessing pressure sensors on adaptable wing structures truly transform robotic flight control methods? Scientists at Tokyo’s Institute of Science have achieved a breakthrough in detecting wind routes with unprecedented accuracy, leveraging seven pressure sensors on flapping wings and a sophisticated convolutional neural network model. This groundbreaking discovery, inspired by the unique pressure sensors found in birds and insects, paves the way for innovative advancements in the design and control of flapping-wing aerial robots, enabling them to thrive in a wide range of windy conditions.
Insects and birds feature mechanoreceptors on their wings, which likely collect pressure-sensitive data, aiding in the regulation of their flight. These putative sensors are thought to monitor variations in air flow, body movement, and ambient conditions, enabling adaptive adjustments during flight. Researchers are fascinated by the insect’s capacity for pressure sensing in its wings and are investigating whether biomimetic flapping robots can leverage this capability to gather environmental stream information.
Researchers at the Institute of Science, Tokyo, under the guidance of Affiliate Professor Hiroto Tanaka, published findings on November 11, 2024, detailing their investigation into the application of pressure sensors on hummingbird-inspired wings with versatile flapping capabilities to accurately detect stream instructions during tethered flight in a wind tunnel mimicking hovering conditions under mild wind conditions.
Because of their extreme weight and size constraints, small aerial robots cannot accommodate traditional flow-sensing equipment. Accordingly, employing intuitive wing pressure sensors to instantaneously detect and respond to fluid dynamics without relying on additional equipment would be a valuable innovation, suggests Tanaka.
Researchers connected seven pressure sensors, commonly used in industry due to their affordability and reliability, to a novel wing architecture inspired by the remarkable aerodynamics of hummingbird wings. The wing’s structure has been designed with tapered shafts, reminiscent of natural wings, where each part tapers to support its own unique function, just as a bird’s wing does. A custom-designed flapping mechanism drives the artificial wings, comprising a DC motor, Scotch yoke, and precision gears, creating a smooth oscillation at a frequency of 12 cycles per second. Researchers employed a relatively gentle airflow of 0.8 meters per second within a controlled wind tunnel environment. Pressure on the wing was recorded during flapping exercises conducted under a variety of wind conditions, including seven distinct angles (0°, 15°, 30°, 45°, 60°, 75°, and 90°), as well as one control scenario with no wind. A convolutional neural network (CNN) model was employed for machine learning analysis of the pressure data to classify these wind scenarios.
Watch a close-up examination of the wing mechanism’s movement in our accompanying video, showcasing a slow-motion display of flapping motions under zero-airflow conditions, as well as pressure gauge readings with and without external pressure applied.
Utilizing pressure information and flapping cycle sizes, an exceptional classification accuracy of 99.5% was achieved due to this. Despite a 0.2-fractional reduction in information size, the classification accuracy remarkably persisted at an impressive 85.2%. The single pressure gauge yielded exceptionally high classification accuracy, ranging from 95.2% to 98.8%, when considering a full flapping cycle; however, this performance plummeted to 65.6% or lower when examining rapid 0.2 cycles data? Data suggests that monitoring wing pressure across multiple points enables the precise identification of wind routes within just 0.2 flapping cycles, yielding exceptional accuracy.
By removing the internal wing shafts, the classification accuracy noticeably declined. The diploma obtained a 4.4% average with 0.2 cycle-related data and 0.5% with 1 cycle-related data when all pressure gauges were utilized, correspondingly. When relying on a single pressure gauge, the average accuracy was significantly impacted, with an averaged percentage deviation of 7.2% for 1-cycle data and 6% for 0.2-cycle data. The biomimetic wing shaft structures enhance the wind detection abilities of the wings significantly.
This innovative study furthers our comprehension of how hovering avian and insect species employ subtle pressure sensations from their wing movements to intuitively grasp air currents, thereby enabling them to dynamically adjust their flight patterns. Biomimicry enables the development of innovative flapping-wing aerial robots through the integration of simple yet effective pressure sensors, as Dr. Tanaka observes.
