NRI/Collaborative Research: Robot-Assisted Feeding: Towards Efficient, Safe, and Personalized Caregiving Robots

NRI/合作研究：机器人辅助喂养：迈向高效、安全和个性化的护理机器人

• 批准号: 2132847
• 负责人: Dorsa Sadigh
• 金额: $ 50.85万
• 依托单位: Stanford University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2022
• 资助国家: 美国
• 起止时间: 2022-01-01 至 2025-12-31
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2132847&HistoricalAwards=false
• 关键词: NRI; Collaborative; Research; Robot; Assisted

The goal of this project is to develop caregiving robots that can provide long-term assistance with activities of daily living (ADLs) to people with mobility limitations. Despite great strides taken towards sustainable solutions in controlled environments, robots are far from ready for adoption in real home environments as long-term caregiving solutions. Key factors could be the over-reliance on full autonomy in tasks that require dynamic physical and social interactions in unstructured environments as well as the lack of personalized assistance. Based on the central tenet that robots need to optimize both physical and social interactions to provide efficient, safe, and personalized assistance for ADLs, this work will focus on developing robot-assisted feeding as a long-term caregiving solution for a person with upper-extremity disability in an unstructured, real-home environment. Successful feeding consists of bite acquisition (i.e., picking up a food item) and bite transfer (i.e., moving it into the mouth). This project develops methods to integrate these activities towards the development of an intelligent and personalized robot-assisted feeding system. The models leverage multimodal feedback to develop human-in-the-loop control policies that adapt to a range of human and environmental factors. Realizing that full autonomy can be challenging in unstructured and dynamic environments, the methods will leverage expert human feedback while minimizing the cognitive load and interweave them intelligently with autonomy to arrive at a long-term caregiving solution. This work will have a direct impact on the lives, health, and comfort of millions of people in the world who live with motor impairments. Developing policies that consider the human in the loop at every step and learn from their feedback through multiple modalities will have an impact on many other human-robot interaction domains including but not limited to assistive teleoperation.This project will advance the state of the art of robotics from both a technical and algorithmic perspective. First, novel bite acquisition algorithms will be developed that are capable of picking up deformable objects in unstructured settings. Second, bite-transfer algorithms will be developed that learn from physical feedback provided on the robot or on the utensil when transferring food inside of a person's mouth. Finally, the research team will develop active and adaptive algorithms that tap into other sources of data, such as comparisons or language instructions, to intelligently improve the acquisition and transfer algorithms and personalize the feeding experience. This work leverages the idea of learning from multimodal human feedback---specifically by embracing physical interactions rather than trying to avoid them---to better manipulate and transfer food. The assistive acquisition and transfer algorithms will be extensively evaluated through human subject studies. The algorithms will be implemented on multiple high-degree-of-freedom robotic platforms across labs. Planned user studies and low-level implementations will advance the state of robotics outside of assistive feeding, particularly towards other ADLs or Instrumental ADLs (IADLs) in home settings, such as meal-preparation, cooking, and housework.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.

该项目的目标是开发护理机器人，可以为行动不便的人提供日常生活活动（ADL）的长期帮助。尽管在受控环境中的可持续解决方案方面取得了长足的进步，但机器人还远远没有准备好在真实的家庭环境中作为长期的可持续解决方案。关键因素可能是在非结构化环境中需要动态物理和社会互动的任务中过度依赖完全自主权，以及缺乏个性化的帮助。基于机器人需要优化身体和社会互动以为ADL提供高效，安全和个性化的帮助的中心原则，这项工作将专注于开发机器人辅助喂养作为上肢残疾人在非结构化，真实家庭环境中的长期解决方案。成功的喂养包括咬获取（即，拾取食物）和咬合转移（即，将其放入口中）。该项目开发了将这些活动整合到智能和个性化机器人辅助喂养系统开发中的方法。这些模型利用多模态反馈来开发人在回路控制策略，以适应一系列人为和环境因素。意识到完全自主在非结构化和动态环境中可能具有挑战性，这些方法将利用专家的人类反馈，同时最大限度地减少认知负荷，并将它们与自主智能地交织在一起，以实现长期的自主解决方案。这项工作将对世界上数百万运动障碍患者的生活、健康和舒适度产生直接影响。开发在每一步都考虑人类的策略，并通过多种方式从他们的反馈中学习，这将对许多其他人机交互领域产生影响，包括但不限于辅助遥操作。该项目将从技术和算法的角度推进机器人技术的发展。首先，将开发能够在非结构化环境中拾取可变形物体的新型咬合采集算法。其次，将开发咬转移算法，该算法在将食物转移到人的口中时从机器人或器具上提供的物理反馈中学习。最后，研究团队将开发主动和自适应算法，利用其他数据来源，如比较或语言指令，以智能地改进获取和传输算法，并个性化喂食体验。这项工作利用了从多模态人类反馈中学习的想法-特别是通过拥抱物理交互而不是试图避免它们-以更好地操纵和转移食物。辅助采集和传输算法将通过人类受试者研究进行广泛评估。这些算法将在实验室的多个高自由度机器人平台上实现。计划中的用户研究和低水平实施将推动机器人技术在辅助喂养之外的发展，特别是在家庭环境中的其他ADL或工具性ADL（IADL），如备餐，烹饪和家务。该奖项反映了NSF的法定使命，并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，被认为值得支持。

项目成果

Stabilize to Act: Learning to Coordinate for Bimanual Manipulation

• DOI: 10.48550/arxiv.2309.01087
• 发表时间: 2023-09
• 期刊: ArXiv
• 作者: J. Grannen;Yilin Wu;Brandon Vu;Dorsa Sadigh

In-Mouth Robotic Bite Transfer with Visual and Haptic Sensing

具有视觉和触觉感应的口内机器人咬合转移

• 发表时间: 2023
• 期刊: International Conference on Robotics and Automation (ICRA
• 作者: Shaikewitz, Lorenzo;Wu, Yilin;Belkhale, Suneel;Grannen, Jennifer;Sundaresan, Priya;Sadigh, Dorsa
• 通讯作者: Sadigh, Dorsa

Eliciting Compatible Demonstrations for Multi-Human Imitation Learning

• DOI: 10.48550/arxiv.2210.08073
• 发表时间: 2022-10
• 作者: Kanishk Gandhi;Siddharth Karamcheti;Madeline Liao;Dorsa Sadigh

Gesture-Informed Robot Assistance via Foundation Model

通过基础模型提供手势通知机器人辅助

• 发表时间: 2023
• 期刊: Proceedings of the 7th Conference on Robot Learning (CoRL
• 作者: Lin, Li-Heng;Cui, Yuchen;Hao, Yilun;Xia, Fei;Sadigh, Dorsa
• 通讯作者: Sadigh, Dorsa

Assistive Teaching of Motor Control Tasks to Humans

人类运动控制任务的辅助教学

• 发表时间: 2022
• 期刊: Conference on Neural Information Processing Systems (NeurIPS
• 作者: Megha Srivastava, Erdem Bıyık