The Role of the Dynamics of Relative Motion in Information Passing in Natural and Engineered Collective Motion

Abstract

The breathtaking motions of natural groups such as bird flocks and

fish schools have been a major inspiration for research in the

developement of control algorithms for multi-robotic systems. A

main thrust of research in this area is to understand how individual

behaviors result in group behaviors. In this dissertation, we are

concerned with how individuals --- either natural or engineered ---

move dynamically relative to one another and how this affects

information passing in the group, thereby impacting the performance

of the group when completing tasks.

The work presented in this dissertation is inspired by observations

of coordinated speed oscillations in schools of banded killifish

(Fundulus diaphanus). We analyze trajectory data of

schools of two and three killifish and show that their speed

oscillations produce periodic relative motion that modulates

line-of-sight visibility between fish. We leverage tools from graph

theory and dynamical models of consensus decision making to show

that this phenomenon can significantly improve group decision-making

performance.

We also investigate the potential benefits dynamic relative motion

for engineered systems, particularly mobile sensor networks. We

investigate the ways that relative motion can enhance mobile sensor

networks with respect to group connectivity and decision-making

performance. We also derive control laws to stabilize groups of

mobile agents that are engaged in killifish-like coordinated speed

oscillations into a rich family of formations along circular

trajectories. We present design guidelines for these types of

formations and algorithms to compute what speed oscillation waveform

will produce or approximate a formation with a given shape.

Motivated by the desire to perform controlled experiments that can

further illuminate these and other themes related to collective

motion in animal groups, we present the design of a testbed in which

a robotic fish interacts with a school of live fish in real time.

Each robotic fish consists of a wheeled robot and a model fish; the

model fish is moved about a shallow tank of water via magnetic

coupling to a wheeled robot beneath the tank. A significant part of

the design of this testbed is the real-time computer vision tracking

of the robotic fish and the school of fish. The technology that we

have developed for the robotic fish testbed is relevant to a number

of other applications.