Designing and Implementing Strategies for Human-Robot Interaction Using a Nonlinear Model of Opinion Dynamics

Abstract

Understanding inter-agent interactions is a pivotal component in implementing robust multi-robot systems. In this context, game-theoretic approaches have gained significant traction because of their ability to represent how multiple agents can have different incentives for different strategies. However, commonly used solution concepts like the Nash equilibrium often do not accurately represent real-world interactions between decision makers. Networked groups in reality are able to break deadlocks and make decisions that are often not predicted based on rational individual decisions. For example, the Nash equilibrium in the prisoner’s dilemma game predicts that agents will not cooperate, but human/animal agents in fact often do choose to cooperate when engaged in repeated game settings resembling the idealized prisoner's dilemma model. These cooperating behaviors have inspired the use of a framework based on these games for the purpose of human-robot coordination, where the interaction between agents is modeled through a nonlinear model of opinion dynamics. Previous work has investigated this model of opinion dynamics for understanding rationality and reciprocity in iterated games like the prisoner’s dilemma and the stag hunt game. In this project, we use this nonlinear opinion dynamics model to implement strategies for robust human-robot interaction, where the cooperation between humans and robots that arises is based on coordination games. We present a theoretical analysis of this framework and conduct a preliminary robotic implementation of this coordination strategy using a game designed for a driving application.