NRI: INT: Learning-Enabled Robot Support of Daily Activities for Successful Activity Completion

人口正在老龄化——预计到2050年，85岁以上的人口数量将增加两倍。即使在我们目前的人口中，估计有9%的65岁以上的成年人和50%的85岁以上的成年人在日常活动中需要帮助，美国每年的费用大约是2万亿美元。考虑到经济和生活质量的成本，我们迫切需要更好地利用智能技术，使个人能够在自己的家中独立生活，从而帮助个人和整个社会。这项提议的工作将创造一种新的多智能体机器人系统，称为机器人活动支持（RAS），为老年人和其他需要帮助独立完成日常生活中常见活动的人提供家庭活动支持。该系统将依靠智能家居和移动机器人之间的合作来学习个人的活动习惯。RAS将利用这些信息提供活动干预，帮助智能家居居民开始并成功完成重要的日常活动，提高功能独立性。该系统将代表移动机器人、具有多个异构传感器的智能家居代理和具有不同角色的多个人类之间的协作，而不是探索具有多个相同平台的协作机器人系统。对于这次合作，项目团队将使用一个定制的机器人，它将与团队的CASAS智能家居架构合作。RAS将整合护理人员在循环中的主动学习来改进其模型。该团队将使用迭代的以用户为中心的开发过程来增强移动机器人平台的设计。RAS系统将利用居民和护理人员的主动学习来学习共同的活动以及如何支持这些活动。例如，如果居民没有在正常时间开始任务，机器人可能会提示她吃早餐，提醒居民谷物在哪里，如果需要机器人无法提供的帮助，机器人会通知附近的护理人员。该团队将根据历史智能家居数据（来自他们的100多个部署）、校园智能家居、有健康老年人照顾者的家庭以及表现出认知限制的老年人的家庭评估RAS。

The population is aging - the estimated number of individuals over the age of 85 will triple by 2050. Even in our current population an estimated 9% of adults age 65+ and 50% of adults age 85+ need assistance with everyday activities, and the annual cost for the United States is roughly $2 trillion. Given the economic and quality of life costs, there is a critical need to better use smart technologies so that individuals can live independently in their own homes, helping both individuals and society as a whole. This proposed work will create a novel multi-agent robot system, called Robotic Activity Support (RAS), that provides in-home activity support for older adults and others that need assistance to independently perform common activities of daily living. The system will rely on cooperation between a smart home and a mobile robot to learn activity routines for an individual. RAS will use this information to provide activity interventions that help smart home residents initiate and successfully complete important daily activities and improve functional independence. Rather than explore co-robot systems with multiple identical platforms, the system will represent a collaboration between a mobile robot, a smart home agent with multiple heterogeneous sensors and multiple humans with distinct roles. For this collaboration, the project team will employ a custom robot that will partner with the team's CASAS smart home architecture. RAS will incorporate caregiver-in-the-loop active learning to improve its models. the team will use an iterative user-centered development process to enhance the mobile robotic platform design. The RAS system will use active learning from both the resident and caregiver to learn common activities and how it can support such activities. For example, the robot may prompt a resident to eat breakfast if she does not initiate the task at the normal time, remind the resident where the cereal is, and notify a nearby caregiver if help is needed that is beyond the robot's capabilities. The team will evaluate RAS on historic smart home data (from their more than 100 deployments), in their on-campus smart home, and in a home with a healthy older adult caregiver, as well as an older adult exhibiting cognitive limitations.