NRI: FND: Using Template Models to Identify Exoskeleton User Intent

外骨骼对于因脊髓损伤(SCI)和中风等神经损伤而导致的运动障碍患者具有恢复活动能力和提高生活质量的巨大潜力。几年来，临床外骨骼的使用加速了步态再训练，使患者能够在每次训练中走更多的步，练习更可重复的步态模式，并更密切地监测他们的进展。如今，这项技术已经成熟到了多种获得FDA批准的系统都可以商业化的程度。然而，无论是在临床上还是在个人使用中，这项技术都需要更广泛地采用外骨骼，才能充分发挥其影响日常生活的潜力。这种采用的关键障碍是缺乏一个流畅和透明的界面来确定用户打算如何与外骨骼一起移动。该项目寻求通过开发一种可靠的方法来确定用户的意图，从而使外骨骼能够在日常生活的不同场景中使用。由于步行的基本机制是真正通用的，因此方法是利用相对简单的步行模板模型，这些模板模型将这些机制编码到算法中，从而更可靠地识别用户意图。有趣的活动包括从休息开始，改变步行速度、方向和节奏，爬楼梯/下楼梯，以及停下来。这些模型将应用于健康受试者和脊髓损伤患者，以确定共同点和关键差异。除了对外骨骼终端用户的好处外，该项目还将促进与K-12教师和中学生的接触，通过强调工程如何直接提高生活质量来促进STEM学科的教育。该项目旨在通过使用简单的运动模板模型来改进下半身外骨骼的人机界面技术，以更有效地捕捉用户意图。在大多数现有的方法中，将传感器数据映射到用户的状态/活动被视为黑盒模式识别问题。相反，降阶模型或模板是低维动力系统，它捕捉到了行走的基本机制，在此基础上表现出更复杂的行为。该项目将研究基于物理学的模板模型，以增强外骨骼用户对名义行走步态(速度、方向和节奏)的微小变化意图的推断。利用仿生模板控制算法，扩展卡尔曼滤波和非参数贝叶斯方法将被用来解决随机意图观测器问题。下一步，这项工作将扩展到检测用户意图的转变，包括开始、停止和步态进入/离开楼梯。在整个过程中，平行分析将研究方法的定制，包括为有运动障碍的外骨骼用户使用完全不同的模板模型。数据驱动的Floquet分析将应用于来自健康用户的数据转换，以增强脊髓损伤用户的意图识别。新意图识别的实验将被纳入FDA批准的外骨骼的控制系统中，该外骨骼将用于评估相对于现有技术的识别延迟和推广性能。这些实验将首先在健康受试者身上进行，然后在不完全脊髓损伤的个人身上进行。最终，该项目有可能通过更有效地捕获用户意图来提高外骨骼在康复、搜救、军事和工业应用领域的影响。

Exoskeletons have great potential to restore mobility and improve quality of life for individuals with locomotor impairments due to neurotrauma such as spinal cord injury (SCI) and stroke. For several years, clinical exoskeleton use has accelerated gait retraining by enabling patients to take more steps per session, practice more repeatable gait patterns, and have their progress more closely monitored. Today, the technology has matured to the point that multiple systems with FDA approval are commercially available. Still, broader exoskeleton adoption, both in the clinic and for personal use, is needed for the technology to achieve its full potential to impact daily life. The key barrier to such adoption is the lack of a fluent and transparent interface to determine how the user intends to move in conjunction with the exoskeleton. This project seeks to enable exoskeleton use in the diverse scenarios of daily life by developing a robust approach to determine a user's intent. Since the fundamental mechanics of walking are truly universal, the approach is to leverage relatively simple template models of walking that encode these mechanics within algorithms that more reliably identify user intent. Activities of interest include starting from rest, changing walking speed, direction, and cadence, stair ascent/descent, and stopping. The models will be applied to both healthy subjects and individuals with spinal cord injury to identify commonality and critical differences. Beyond the benefits to exoskeleton end-users, the project will facilitate outreach to both K-12 teachers and middle school students to promote education in the STEM disciplines by highlighting how engineering can directly improve quality of life. This project aims to improve human machine interface technologies for lower-body exoskeletons by using simple template models of locomotion to more effectively capture user intent. In most existing approaches, mapping sensor data to the user's state/activity is treated as a black-box pattern recognition problem. In contrast, reduced order models or templates are low-dimensional dynamical systems that capture the fundamental mechanics of walking, upon which more complex behaviors play out. The project will investigate physics-based template models to augment inference of exoskeleton user intent for small changes to nominal walking gait (speed, direction, and cadence). Leveraging bio-inspired template control algorithms, extended Kalman filtering and non-parameteric Bayesian approaches will be investigated to solve a stochastic intent observer problem. Next, the work will be extended to detect transitions in user intent for starting, stopping, and gait progression to/from stairs. Throughout, parallel analysis will study customization of the methodology, including the use of entirely different template models, for exoskeleton users with locomotor impairments. Data-driven Floquet analysis will be applied for translation of data from healthy users to augment intent recognition in users with spinal cord injury. Experiments with the new intent recognition will be incorporated into the control system of an FDA-approved exoskeleton that will be used to assess recognition delay and generalization performance of the methods relative to existing techniques. These experiments will be conducted first with healthy subjects and subsequently with individuals with incomplete spinal cord injury. Ultimately, the project has the potential to improve the impact of exoskeletons in areas spanning rehabilitation, search-and-rescue, military, and industrial applications by more effectively capturing user intent.