Collision-free Navigation with Unknown and Heterogeneous Social Preferences

Abstract

In this study, we address cooperative navigation using multiple agents with social preferences such as egoism and altruism. Previous studies have proposed collision-free navigation with social preferences under the assumption that the social preferences of other agents are available. However, if some agents cannot express their social preferences to others, this method is unavailable. In this study, we consider the cooperative navigation problem where agents do not know the social preference of the others. Our social preference estimation scheme can realize provably collision-free navigation. Agents estimate social preferences based on their trajectories. Then, they modify existing Voronoi cells by using their positions and the estimated social preferences of other agents. Moving within the modified Voronoi cells enables all agents to avoid collision with others, even if the actual social preferences are unknown. Through a simulation study for various distributions of social preferences, we show that the proposed method can avoid collision and achieve navigation performance close to that of the existing method using the actual social preferences.