In-Situ Learning from a Domain Expert for Real World Socially Assistive Robot Deployment
In-Situ Learning from a Domain Expert for Real World Socially Assistive Robot Deployment
复制标题
向现实世界的社交辅助机器人部署领域专家进行现场学习
DOI：
10.15607/rss.2020.xvi.059
复制
发表时间：
2020
期刊：
影响因子：
--
通讯作者：
U. Leonards
中科院分区：
文献类型：
--
作者：
Katie Winkle;S. Lemaignan;P. Caleb;Paul A. Bremner;A. Turton;U. Leonards
社交辅助机器人（SAR）的有效性取决于它们激励特定用户行为的能力，例如参与任务，需要根据用户的需求和动机定制复杂的社交互动。来自以人为本的领域（如医疗保健）的专业人士是此类交互的专家，但他们对SAR开发的贡献传统上仅限于识别应用和关键设计要求。在这项工作中，我们展示了交互式机器学习如何为这些专家提供一种参与机器人设计和自动化的每个阶段的方法，以及采用这种方法的价值。我们提出了一种新的技术框架，用于原位，在线交互式机器学习，可用于生态有效的人机交互。使用这个框架，我们能够在辅助机器人的高维应用中生成完全自主，适当和个性化的机器人行为。
—The effectiveness of Socially Assistive Robots (SAR) relies on their ability to motivate particular user behaviours, e.g. engagement with a task, requiring complex social interactions tailored to the needs and motivations of the user. Professionals from human-centred domains such as healthcare are experts in such interactions, but their ability to contribute to SAR development has traditionally been limited to the identification of applications and key design requirements. In this work we demonstrate how interactive machine learning offers a way for such experts to be involved at every stage of design and automation of a robot, as well as the value of taking this approach. We present a novel technical framework for in-situ, online interactive machine learning that can be used in ecologically-valid human-robot interactions. Using this framework, we were able generate fully autonomous, appropriate and personalised robot behaviour in a high-dimensional application of assistive robotics.