Bio-Inspired Vision-Based Leader-Follower Formation Flying in the Presence of Delays

Abstract

Flocking starlings at dusk are known for the mesmerizing and intricate shapes they generate, as well as how fluid these shapes change. They seem to do this effortlessly. Real-life vision-based flocking has not been achieved in micro-UAVs (micro Unmanned Aerial Vehicles) to date. Towards this goal, we make three contributions in this paper: (i) we used a computational approach to develop a bio-inspired architecture for vision-based Leader-Follower formation flying on two micro-UAVs. We believe that the minimal computational cost of the resulting algorithm makes it suitable for object detection and tracking during high-speed flocking; (ii) we show that provided delays in the control loop of a micro-UAV are below a critical value, Kalman filter-based estimation algorithms are not required to achieve Leader-Follower formation flying; (iii) unlike previous approaches, we do not use external observers, such as GPS signals or synchronized communication with flock members. These three contributions could be useful in achieving vision-based flocking in GPS-denied environments on computationally-limited agents.