Mechanisms and Opportunities for the Design of Robust and Flexible Collective Behavior in Dynamic Multi-Agent Systems with Interactions

Abstract

This dissertation examines the role of feedback and agent level adaptations in the

emergence of collective behavior in multi-agent systems that is robust to perturbations

and flexible in response to different environmental conditions. In particular, we

consider systems with agents that stochastically interact with one another and with

their environment, and study how adaptations at the agent level in response to these

interactions lead to the emergence of complex group behavior that evolves in time.

Motivated by the remarkable collective behaviors of animal groups, we study the

regulation of foraging in harvester ants. We use field experiments and mathematical

modeling to examine how interactions between incoming foragers carrying food and

available foragers inside the nest yield foraging rates that are robust to uncertainty

and responsive to temperature and humidity across minute-to-hour timescales. We

show that feedback from outgoing foragers returning to the nest generates stable

colony foraging rates and propose that foragers modify their susceptibility to interactions

after they become exposed to the environment to explain how the foraging

rates adjust to temperature and humidity.

We then examine the role of reinfection and changes in susceptibility in contagion

processes that propagate through local interactions. We study a model for reinfection

called Susceptible-Infected-Recovered-Infected (SIRI) in which the susceptibility

of individuals changes irreversibly after a first exposure to the infection. We show

that in both well-mixed and network topologies, the transient and steady-state group

dynamics are characterized by two critical system parameters that capture the sensitivity

of the group to stimulus before and after the adjustments in susceptibility. We

obtain analytical results that yield exact predictions on the effect of network structure

and individual behavior on group outcomes.

Our results suggest that agent level adaptations in susceptibility, in combination

with feedback across different timescales, are a general principle for robust and

flexible collective behavior. This presents mechanisms and opportunities for the study,

design, and control of multi-agent systems with cohesive group behaviors that adapt

to different conditions.