Measuring collaborative emergent behavior in multi-agent reinforcement learning

TL;DR

This paper introduces a new quantitative metric to assess collaboration in multi-agent reinforcement learning, addressing the gap between performance and genuine collaborative behavior, with potential applications in human-agent teaming.

Contribution

The paper presents a novel metric for measuring collaboration in multi-agent RL, which can improve understanding and training of collaborative behaviors.

Findings

The metric effectively distinguishes between performance and true collaboration.

It provides a quantitative tool for evaluating multi-agent cooperation.

Potential to enhance training signals for collaborative behaviors in RL.

Abstract

Multi-agent reinforcement learning (RL) has important implications for the future of human-agent teaming. We show that improved performance with multi-agent RL is not a guarantee of the collaborative behavior thought to be important for solving multi-agent tasks. To address this, we present a novel approach for quantitatively assessing collaboration in continuous spatial tasks with multi-agent RL. Such a metric is useful for measuring collaboration between computational agents and may serve as a training signal for collaboration in future RL paradigms involving humans.