CAREER: Not all Collisions are Bad: Leveraging Physical Interactions towards User-Empowered Robotic Caregiving

为长期护理建立安全、高效和有意义的机器人个人辅助是机器人学的一个重大挑战。2014年，超过四分之一的美国人患有残疾，2420万18岁或18岁以上的人在日常生活活动中需要帮助，如进食、洗澡、转移、穿衣、厕所和走动。看护机器人有可能帮助增加或延长用户独立性，同时减轻看护人员的负担。然而，目前几乎没有任何护理机器人为执行日常生活活动提供长期护理。其中一个主要原因是机器人对身体接触的基本限制，而身体接触是进行这些活动所必需的。因此，我们的关键洞察力是利用这种身体接触，而不是避免它们，并在赋予用户权力的同时学习安全和高效的身体互动。该学院早期职业发展(CALEAR)项目专注于学习机器人控制政策，这些政策利用机器人辅助喂养和洗床活动中的身体接触，同时赋予用户自己帮助脊髓损伤护理接受者的决策能力。该项目的目标是通过设计利用各种多模式物理互动的控制政策来安全有效地执行日常生活活动(ADL)，发展重塑机器人物理护理领域的基本能力。这还需要通过授权用户在不完全依赖完全但脆弱的自主性的情况下做出关键决策，来解决非结构化环境中的感知和行动不确定性。辅助机器人和人与机器人的物理交互领域主要关注在末端执行器使用精心控制的动作进行有意接触，这些动作是为传感而优化的。这不包括可能需要对偶发和/或分布式接触进行推理的物理护理的现实场景。该项目旨在弥补这一差距，并朝着将机器人物理护理的实验室研究转化为现实世界的部署迈出了第一步。该研究计划分为三个重点，第一个重点是设计自主控制策略，使用多模式感觉输入进行安全和高效的机器人物理护理。第二个推力将用户带入循环，以增强自主控制策略，并通过将用户纳入关键决策过程来增强用户的能力。最后，第三个重点是设计能够适应用户不断变化的认知工作量的机器人护理控制策略。该项目专注于机器人辅助喂养和洗床的活动，旨在评估这些方法，让那些在真实家中患有C1-C4脊髓损伤的人获得长期帮助。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。

Building safe, efficient, and meaningful robotic individual assistance for long-term caregiving is a grand challenge of robotics. More than a quarter of people living in the United States in 2014 had a disability, and 24.2 million people aged 18 or older required assistance with activities of daily living such as feeding, bathing, transferring, dressing, toileting, and ambulating. Caregiving robots have the potential to help increase or prolong user independence while reducing caregiver burden. However, there are hardly any caregiving robots that provide long-term care for performing activities of daily living, currently. One of the primary reasons is the fundamental limitation to which robots can reason about physical contact which is essential to perform these activities. Therefore, our key insight is to leverage this physical contact instead of avoiding them and to learn safe and efficient physical interactions while empowering the users. This Faculty Early Career Development (CAREER) project focuses on learning robot control policies that leverage physical contact during the activities of robot-assisted feeding and bed-bathing while empowering the users themselves with decision-making capabilities for assisting care recipients with Spinal Cord Injury.The goal of this project is to develop fundamental capabilities towards reshaping the field of robotic physical caregiving by designing control policies that leverage varied multimodal physical interactions to perform activities of daily living (ADL) safely and efficiently. This also entails addressing perception and action uncertainty in unstructured environments by empowering the users to make critical decisions without completely relying on full but fragile autonomy. The fields of assistive robotics and physical human-robot interaction have primarily focused on deliberate contact at the end-effector using carefully controlled actions that are optimized for sensing. This does not capture the realistic scenarios of physical caregiving that may require reasoning about incidental and/or distributed contacts. This project aims to cover this gap and takes the first steps towards translating laboratory research in robotic physical caregiving to real-world deployment. The research plan is organized into three thrusts where the first thrust focuses on designing autonomous control policies that use multimodal sensory inputs for safe and efficient robotic physical caregiving. The second thrust brings users in the loop to augment autonomous control policies and empowers the users by including them in the critical decision-making process. Finally, the third thrust focuses on designing control policies for robotic caregiving that can adapt to a user's changing cognitive workload. This project focuses on the activities of robot-assisted feeding and bed-bathing and aims to evaluate these methods with individuals with C1-C4 Spinal Cord Injury in their real homes towards long-term assistance.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.