EAGER: Hierarchical Contrastive Explanations for Robot-Human Communication

智能辅助机器人在各行各业都有改善我们社会的潜力：它们可以帮助我们照顾老人和病人，在紧急情况下充当第一反应者，并启动行星外探索。然而，今天的机器人需要训练有素的专家来进行定制、配置和维修。这不仅使辅助机器人在社会中的潜在好处难以实现，还给数百万劳动力的未来就业带来了很大的不确定性。为了帮助解决这些问题，该项目开发了新的方法，让机器人考虑到用户的熟练程度，向人类解释它们的行为。因此，机器人将能够根据人们已经了解的机器人的能力和限制来定制它们的解释。该项目专注于自动向用户解释意外的机器人行为，而用户对潜在的任务和/或机器人的知识可能不准确。这可以用于有效地诊断规范问题，并针对所需的行为定制机器人。提出的方法形式化了三个一般原则：1)根据观众定制解释；2)将解释视为一个交互过程；3)使用向机器人提出的问题来估计用户的专业水平。在该框架中，用户可以呈现S/他认为更自然的替代机器人行为的陪衬或反事实建议。所提出的方法使用抽象模型的格子来估计用户的熟练程度，并使用最少的附加细节来计算所提出的备选方案不起作用的原因。通过这种方式，系统可以产生与用户的熟练程度相对比和一致的解释。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Intelligent assistive robots have the potential to improve our society in many walks of life: they could help us take care of the elderly and the sick, act as first responders in emergencies, and kickstart extra-planetary exploration. However, today's robots require highly trained experts for their customization, configuration, and repair. This not only makes it difficult to realize the potential benefits of assistive robots in society, but also creates large uncertainties in the future of employment for millions in the workforce. To help address such issues, this project develops new ways for robots to explain their actions to humans, considering the proficiency of the users. Thus, robots will be able to tailor their explanations to what someone already understands about the robots' capabilities and limitations.The project focuses on automatically explaining unexpected robot behavior to users with potentially imprecise knowledge about the underlying task and/or the robot. This can be used to efficiently diagnose specification problems and customize robots towards desired behaviors. The proposed approach formalizes three general principles: 1) customizing explanations according to the audience; 2) treating explanation as an interactive process; and 3) using the questions asked of a robot to estimate the user's level of expertise. In this framework, a user may present a foil, or a counterfactual proposal of alternative robot behavior, that s/he finds more natural. The proposed approach estimates the user's proficiency using a lattice of abstract models and computes reasons why the proposed alternatives would not work using minimal additional detail. In this way, the system can produce explanations that are contrastive and aligned with the proficiency of the user.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.