📄 中文摘要
互适应是人机团队中的一个核心挑战，因人类会自然地根据机器人的策略调整其行动。现有方法旨在通过提高训练伙伴的多样性来近似人类行为，但这些伙伴是静态的，无法捕捉人类的适应性行为。让机器人接触适应性行为至关重要。然而，当两个智能体在多智能体环境中同时学习时，它们往往会收敛到不透明的隐式协调策略，这些策略仅适用于与其共同训练的智能体，导致在与新伙伴配对时缺乏泛化能力。为了捕捉人类的适应性行为，研究将人机团队场景建模为一个互动部分可观测的环境。
人机团队中的互适应嵌套训练
出处: Nested Training for Mutual Adaptation in Human-AI Teaming
发布: 2026年2月23日
📄 English Summary
Nested Training for Mutual Adaptation in Human-AI Teaming
Mutual adaptation poses a significant challenge in human-AI teaming, as humans instinctively modify their strategies in response to a robot's policy. Existing methods focus on enhancing the diversity of training partners to mimic human behavior; however, these partners are static and fail to reflect the adaptive nature of humans. Exposing robots to adaptive behaviors is essential. Yet, when both agents learn concurrently in a multi-agent setting, they often converge to opaque implicit coordination strategies that are only effective with the agents they were co-trained with, leading to a lack of generalization when paired with new partners. To effectively capture the adaptive behavior of humans, the study models the human-robot teaming scenario as an Interactive Partially Observable environment.