Individual and Team Trust Preferences for Robotic Swarm Behaviors

TL;DR

This paper introduces a method to analyze human trust preferences in robotic swarms by modeling individual and group preferences through optimization and clustering, aiding in understanding and improving human-swarm interactions.

Contribution

It presents a novel framework combining preference graphs, feature space mapping, and sparse optimization to analyze and cluster human trust preferences for robotic swarm behaviors.

Findings

Preferences can be modeled using linear optimization in feature space.

Trust profiles can be clustered to identify similar individuals.

Group cohesion insights are obtained from unlabeled preference data.

Abstract

Trust between humans and multi-agent robotic swarms may be analyzed using human preferences. These preferences are expressed by an individual as a sequence of ordered comparisons between pairs of swarm behaviors. An individual's preference graph can be formed from this sequence. In addition, swarm behaviors may be mapped to a feature vector space. We formulate a linear optimization problem to locate a trusted behavior in the feature space. Extending to human teams, we define a novel distinctiveness metric using a sparse optimization formulation to cluster similar individuals from a collection of individuals' labeled pairwise preferences. The case of anonymized unlabeled pairwise preferences is also examined to find the average trusted behavior and minimum covariance bound, providing insights into group cohesion. A user study was conducted, with results suggesting that individuals with similar trust profiles can be clustered to facilitate human-swarm teaming.