论文标题
基于对人类代理临时团队的信任来调整行为
Adapting Behaviour Based On Trust In Human-Agent Ad Hoc Teamwork
论文作者
论文摘要
这项工作提出了一个框架，该框架将信任与人类代理团队的临时团队合作场景结合在一起，在该团队中，代理必须与人类合作执行任务。在任务期间，代理必须通过互动和观察来推断人类信任的程度并适应其行为以最大程度地提高团队的表现。为了实现这一目标，我们建议从实验中的人参与者那里收集数据，以定义不同的设置（基于信任级别），并为每个人学习最佳策略。然后，我们创建一个模块来推断当前设置（取决于信任量）。最后，我们在现实世界中验证该框架，并分析这种适应性行为如何影响信任。
This work proposes a framework that incorporates trust in an ad hoc teamwork scenario with human-agent teams, where an agent must collaborate with a human to perform a task. During the task, the agent must infer, through interactions and observations, how much the human trusts it and adapt its behaviour to maximize the team's performance. To achieve this, we propose collecting data from human participants in experiments to define different settings (based on trust levels) and learning optimal policies for each of them. Then, we create a module to infer the current setting (depending on the amount of trust). Finally, we validate this framework in a real-world scenario and analyse how this adaptable behaviour affects trust.
