Journals

...

Gildas Carlin

Landing a helicopter on a moving ship is an error-prone operation, requiring the expertise of highly trained pilots, and benefits greatly from advanced technologies to improve decision-making and action regulation. (Thomas et al. 2021). In stark contrast, landing to forage on a flower looks to be performed with remarkable ease and under visual control in bees (Lehrer and Srinivasan 1992; de Vries et al 2025). Understanding the perceptual behavior of pilots in comparison to that of honeybees is a key driver of innovation in the design of bio-inspired robotic systems. Understanding the mechanisms behind this ability requires accurately tracking the bee’s movement during the landing task. Vision-based tracking techniques using convolutional neural networks (CNNs), such as You Only Look Once (YOLO) algorithm (Redmon et al. 2016), can be specifically trained to detect and track the honeybee-flower system. Our goal is to accurately reconstruct a known trajectory, using a fake bee as a target, with a mean tracking error less than half a centimeter, or about half the length of a honeybee (13 mm).

Nicolas Salvage