NRI: INT: COLLAB: Accelerating Large-Scale Adoption of Robotic Lower-Limb Prostheses through Personalized Prosthesis Controller Adaptation

经历下肢截肢是一个改变生活的事件。 下肢截肢的人严重依赖假腿来支撑他/她自己并保持移动的。 然而，目前使用的大多数腿部假体是无动力的，这会造成很大的不便。 例如，这些无动力的假肢不能推动截肢者使用者向前行走，也不能在站立或爬楼梯时将他们推起来。 随着医疗机器人技术的进步，动力机器人腿假体开始变得越来越普遍。 这些动力假肢可以帮助截肢者更轻松、更自然地行走，并提供动力来支持现有无动力假肢以前无法实现或极其困难的活动（如站立和爬楼梯）。 另一方面，这些机器人假肢比现有的无动力假肢复杂得多，因此调整这种假肢以适应个人用户是非常具有挑战性和耗时的，需要在很长一段时间内与临床医生进行多次办公室访问。 这是动力腿假肢在截肢者中没有得到广泛使用的主要原因之一。在本项目中，研究小组将通过开发一种新的方法来帮助解决这个问题，以自动调整假肢过程。 其主要思想是使用一个类似吊坠的可穿戴传感器来测量上半身的运动，这提供了关于假肢控制程度的丰富信息，以帮助用户行走和执行日常生活中的其他基本任务。 新方法将使用这些信息，并随着时间的推移逐渐调整控制器参数，而无需持续的临床监督。 为了开发这种方法，研究人员将研究经验丰富的假肢专家如何调整假肢参数，并开发一种算法来模拟这一过程。 此外，上半身的运动将被用于更精确地推断截肢者使用者的意图，因此即使他/她在学习使用机器人假肢的过程中改变了运动模式，假肢也可以可靠地理解用户想要做什么。通过在该项目中进行研究，研究人员旨在开发一个完整的个性化假肢控制器自适应（PPCA）系统，其在两个级别上提供个性化控制器适配：1）自动运动控制器调谐，以及2）自动意图识别器适配。 研究人员预计将在相关科学领域做出重大贡献，包括：1）一种新型的可穿戴传感器，它结合了惯性测量单元（IMU），电容传感（用于传感器-躯干相对运动），以及先进的信号处理以提供可靠的躯干运动信息; 2）对机器人假肢辅助截肢者运动的基本了解，以及如何基于人类专业知识的调整优化其步态质量; 3）一种新的基于分类器的意图识别器的准监督自适应，它提供了传统监督学习的优点，同时避免了其主要弱点（在适应不断变化的人类条件方面非常有效，并且不需要重复的训练会话或人类进行的数据标记）。 该项目的影响也将通过其各种教育活动产生，包括通过著名的假肢和矫形教育计划向未来的假肢临床医生介绍机器人技术，以及在本科研究中创建动手机器人项目，这也可以作为K-12推广的重要工具，以吸引不同年龄组的儿童进入科学和工程领域。

Experiencing a lower limb amputation is a life-changing event. A person with lower-limb amputation relies heavily on a leg prosthesis to support himself/herself and stay mobile. However, most leg prostheses in current use are unpowered, which can cause great inconvenience. For example, these unpowered prostheses cannot propel the amputee users forward in walking, and cannot push them up while standing up or climbing stairs. With the advances in medical robotic technologies, powered robotic leg prostheses are starting to become more common. These powered prostheses can help amputees to walk more easily and naturally, and also provide power to support activities previously unattainable or extremely difficult with existing unpowered prostheses (such as standing up and stair climbing). On the other hand, these robotic prostheses are much more complex than the existing unpowered prostheses, so tuning such a prosthesis to fit an individual user is very challenging and time-consuming, requiring numerous office visits with a clinician over a long period of time. This is one of the main reasons why powered leg prostheses have not gained extensive use among the amputee users.In this project, the research team will help to solve this problem by developing a new method to automate the prosthesis tuning process. The main idea is to use a pendant-like wearable sensor to measure upper body motion, which provides rich information about how well the prosthesis is being controlled to help the user to walk and perform other basic tasks of daily life. The new method will use such information and gradually tune the controller parameters over time without the need for constant clinical supervision. To develop this method, the researchers will study how an experienced prosthetist tunes the prosthesis parameters and develop an algorithm to mimic this process. Furthermore, the upper body motion will be used to infer the amputee user's intention more precisely, so the prosthesis can reliably understand what the user intends to do even if he/she changes the motion pattern while learning to use the robotic prosthesis.By conducting research in this project, the researchers aim to develop a complete Personalized Prosthesis Controller Adaptation (PPCA) system, which provides personalized controller adaption on two levels: 1) automatic motion controller tuning, and 2) automatic intent recognizer adaptation. The researchers anticipate to make significant contributions to the related scientific areas, including: 1) a novel wearable sensor that incorporates an inertial measurement unit (IMU), capacitive sensing (for sensor-torso relative motion), and advanced signal processing to provide reliable trunk motion information; 2) fundamental understanding of robotic prosthesis-assisted amputee locomotion, and how human expertise-based tuning optimizes its gait quality; 3) a novel quasi-supervised adaptation of classifier-based intent recognizer, which provides the advantages of the traditional supervised learning while avoiding its major weakness (highly effective in adapting to changing human conditions, and no repeated training sessions or human-conducted data labeling required). Impacts of this project will also be generated by its various education activities, including the introduction of robotic technologies to the future prosthetic clinicians through a renowned prosthetics and orthotics education program, and the creation of hands-on robotic projects in undergraduate research, which can also serve as important tools in the K-12 outreach to attract children at different age groups to the science and engineering fields.