An Empirical Study of Communication-Free Coordination in Human-Robot Teams Through a Coverage Control Game

Abstract

We investigate the performance in coverage control problems, where some robots are controlled by human operators and there are no explicit communications among the robots for coordination. One example of such a scenario is a team of unmanned and manned vehicles together pursuing a surveillance mission, where each vehicle operates based on local observations without communicating with others due to physical or strategic limitations. For such scenarios, there exist distributed algorithms that ensure (near-) optimal long-run average performance when followed by all robots. This paper is focused on how the team performance changes when some robots are controlled by human operators rather than following such an optimal algorithm. For the empirical analysis, we have designed a multi-player computer game, where each player (human operator) controls a single robot and the autonomous robots follow a noisy greedy algorithm to optimize their marginal contribution to the overall coverage. We present the results obtained on multiple maps with a team of four robots, where the number of players range from zero (all robots are autonomous) to four (each robot is controlled by a player). Our results indicate that long-run average performance degrades with the introduction of human players, but this effect is not always monotonous with respect to the number of human players. Furthermore, through post-test questionnaires we showed that performance is a good predictor of the outcome in human subjective assessments. On the other hand, the number of human players in a team was not found to have any significant effect on subjective assessment.