Sign In

Communications of the ACM

ACM TechNews

The Deep Learning Way to Design Fly-Like Robots

View as: Print Mobile App Share:
A concept design of a fly robot.

Researchers have developed a deep-learning-based motion-capture software that uses multiple camera views to model the movements of a fly in three dimensions.

Credit: P. Ramdya/EPFL

Researchers at Switzerland's Swiss Federal Institute of Technology, Lausanne (EPFL) have developed motion-capture software that uses deep learning to model a fruit fly's movements in three dimensions (3D), in the hope of designing fly-like robots.

The DeepFly3D system uses multiple cameras to record images of a fly crawling atop a small floating ball, which the software processes.

DeepFly3D can infer the insect's pose in three dimensions, predicting and calculating behavioral measurements at ultra-fine resolution without the need for manual calibration.

The system also employs active learning to improve its own performance.

EPFL's Pavan Ramdya said, "If we learn how [the fly] does what it does, we can have important impact on robotics and medicine and, perhaps most importantly, we can gain these insights in a relatively short period of time."

From EPFL (Switzerland)
View Full Article


Abstracts Copyright © 2019 SmithBucklin, Washington, DC, USA


No entries found