Dávid Mezey: Minimal Visual Interactions and Collective Movement of Swarming Robots

BCCN Berlin / Technische Universität Berlin

 

Abstract

 

Collective motion of a group of individuals in nature (such as schools of fish or flocks of birds) is a captivating phenomenon that inspires real-world applications of robotic swarms planned to be used in various scenarios in the future. In this work a proof of concept has been presented to demonstrate that a first-of-its-kind purely vision-based flocking algorithm inspired by natural systems is a promising candidate to realize a cohesive collisionless collective motion in artificial robot swarms.

Although numerous other models have been already designed to describe local social interactions between the individuals of animal groups during collective motion, common flaws of these models limit our understanding of sensory information-based collective behavior and make real-life applications less practical. A minimal, purely vision-based framework that provides improvements to such models has been only recently proposed based on observations that simple vision-based local attraction-repulsion interactions can serve as the source of collective motion patterns in fish. Within the framework of this thesis this vision-based algorithm has been wrapped into a self-designed software stack as the controller of physical swarming robot agents with relatively cheap hardware design. The impact of initialization, algorithm parameters and physical constraints were studied qualitatively with realistic robot models using a physical simulation environment. According to these preliminary results a set of parameter settings were chosen to carry out real-life experiments with a minimal number of swarming robot agents and to show that cohesive, collisionless collective motion arises in such an artificial system.

As a result, a possible alternative has been presented to traditional collision avoidance and flocking algorithms in the field of robotics and an experimental model system has been proposed suitable to better understand sensory information-based collective motion in natural swarms.

 

Additional Information

Master Thesis Defense

 

Organized by

Dr. Pawel Romanczuk   & Prof. Verena Hafner   / Lisa Velenosi