CAREER: Democratizing Robot Learning for Assistive Robotics in MCI

职业：MCI 辅助机器人的机器人学习民主化

• 批准号: 2340177
• 负责人: Matthew Gombolay
• 金额: $ 59.98万
• 依托单位: Georgia Tech Research Corporation
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2024
• 资助国家: 美国
• 起止时间: 2024-08-01 至 2029-07-31
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2340177&HistoricalAwards=false
• 关键词: CAREER; Democratizing; Robot; Learning; Assistive

The US population is aging, and 18 percent of adults over 60 years of age have Mild Cognitive Impairment (MCI). Of those with MCI, up to 15 percent develop dementia down the line. Unfortunately, there is an increased shortage of healthcare workers, and the cost of at-home nursing care is prohibitive. To meet this challenge, this Faculty Early Career Development (CAREER) project aims to democratize interactive robot learning to enable care partners (e.g., spouses, children, nurses) to program and personalize a robot's behavior through intuitive modes of interaction to assist the care member (i.e., person with MCI) with activities of daily living. Robot learning is achieved using Learning from Demonstration (LfD), which is about computational methods and interfaces to enable end-users to teach robots new skills through interaction (e.g., skill demonstration). Traditional robotics deployments typically rely on a large number of experts and create robots that are expensive and not easily adapted. Instead, LfD offers a scalable alternative by leveraging efficient algorithms and interactions with non-expert end users. Despite its potential, and decades of research, LfD has not been deployed broadly, in part because these systems do not provide the care partner end-users with insights into the robot's understanding of the world or how the users can be better teachers. This award seeks to overcome the limitations of traditional robotics (such as cost and scale) and modern robot learning approaches (the fact that they are relegated to the laboratory) by enabling robot learning to be accessible and scalable to support aging in place for persons with MCI. The eventual goal is to have a robot that can be used in the home of someone with MCI to help them with their daily tasks, and that the robot can be easily programmed by someone who does not have specific technical training.To achieve these goals, the research team will develop new LfD algorithms and interfaces in partnership with collaborators at the Cognitive Empowerment Program at Emory University, members of the MCI community, clinicians, and researchers at the Georgia Institute of Technology in a transdisciplinary research effort. With the oversight and input from MCI focus groups, the research team will collaboratively execute three research thrusts. First, the team will conduct transdisciplinary research to collect and open-source a first-of-its-kind, multimodal, longitudinal dataset from care partners interacting with a robot with the goal of programming it via LfD to carry out assistive tasks for care members. Second, the team will formulate novel, explainable Artificial Intelligence techniques enabling users to gain a “theory of mind” of the robot, specifically to foster users gaining insight into the behavior of the robot while learning to collaborate via mixed-initiative LfD interactions. Third, the team will develop novel, peer-teaching LfD algorithms that enables the robot learner to develop a theory of mind about an LfD teacher (i.e. the care partner alongside the person with MCI), leverage that insight to tutor a human teacher, and provide explicit feedback for how the teacher can provide better instruction to the robot. The success of the techniques developed will be based upon improving the robot’s performance at specific tasks, reducing the amount of time required for the user to train the robot, and improving user experience in terms of workload, stress, and perceived usability of the robot. The outcome of this research will be open-source datasets, interfaces, and roadmaps guiding researchers in democratizing robots, transitioning LfD from the laboratory to the real world. Finally, the research team will also develop a new educational outreach program in partnership with high school educators to integrate robotics into their classrooms in underserved communities in Georgia.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.

美国人口正在老龄化，60岁以上的成年人中有18%患有轻度认知障碍（MCI）。在MCI患者中，高达15%的人会发展成痴呆症。不幸的是，卫生保健工作者越来越短缺，家庭护理的费用高得令人望而却步。为了迎接这一挑战，这个教师早期职业发展（CAREER）项目旨在使交互式机器人学习民主化，以使护理伙伴（例如，配偶、孩子、护士）通过直观的交互模式来编程和个性化机器人的行为以帮助护理成员（即，MCI患者）的日常生活活动。机器人学习是使用从演示中学习（LfD）来实现的，LfD是关于计算方法和界面，以使最终用户能够通过交互来教授机器人新技能（例如，技能示范）。传统的机器人部署通常依赖于大量的专家，并创建昂贵且不易适应的机器人。相反，LfD通过利用高效的算法和与非专家最终用户的交互提供了一种可扩展的替代方案。尽管LfD具有潜力，并且经过数十年的研究，但它尚未得到广泛部署，部分原因是这些系统无法为护理伙伴最终用户提供有关机器人对世界的理解或用户如何成为更好的教师的见解。该奖项旨在克服传统机器人技术的局限性（如成本和规模）和现代机器人学习方法的局限性（它们被降级到实验室的事实），使机器人学习能够获得和可扩展，以支持MCI患者的就地老化。最终的目标是有一个机器人，可以在MCI患者的家中使用，帮助他们完成日常任务，并且机器人可以很容易地由没有特定技术培训的人编程。为了实现这些目标，研究团队将与埃默里大学认知授权项目的合作者，MCI社区的成员，临床医生和格鲁吉亚理工学院的研究人员进行了一项跨学科的研究。在MCI焦点小组的监督和投入下，研究团队将协同执行三个研究重点。首先，该团队将进行跨学科研究，收集并开源来自护理合作伙伴的首个多模式纵向数据集，该数据集与机器人进行交互，目标是通过LfD对其进行编程，为护理成员执行辅助任务。其次，该团队将制定新颖的，可解释的人工智能技术，使用户能够获得机器人的“心理理论”，特别是培养用户深入了解机器人的行为，同时学习通过混合主动LfD交互进行协作。第三，该团队将开发新颖的同伴教学LfD算法，使机器人学习者能够开发关于LfD教师（即MCI患者旁边的护理伙伴）的心理理论，利用这种洞察力来指导人类教师，并为教师如何为机器人提供更好的指导提供明确的反馈。开发的技术的成功将基于提高机器人在特定任务中的性能，减少用户训练机器人所需的时间，并改善用户在工作量，压力和机器人可用性方面的体验。这项研究的成果将是开源数据集、接口和路线图，指导研究人员使机器人民主化，将LfD从实验室过渡到真实的世界。最后，研究团队还将与高中教育工作者合作开发一个新的教育推广计划，将机器人技术融入格鲁吉亚服务不足社区的课堂。该奖项反映了NSF的法定使命，并被认为值得通过使用基金会的智力价值和更广泛的影响审查标准进行评估来支持。