NRI: INT: Individualized Co-Robotics

该项目探索了新的方式，以满足个性化患者和用户的需求，定制的主动假肢，电动脚踝和膝盖支撑，以及其他形式的辅助机器人。这类设备的有效性取决于适应人与人之间的差异，如身体类型、行走步态和障碍。一项初步研究调查了一个单独适应的踝关节外骨骼控制模式序列--随着受试者使用该设备而进行修改--可以在多大程度上降低行走的新陈代谢成本。这一结果表明，与未优化的设备相比，分别为每个受试者优化辅助控制模式极大地提高了平均收益，并优于任何先前记录的方法。该项目以这些初步成果为基础，为各种不同情况下的各种辅助技术找到更好和更强大的定制方法。这种人在环中优化人类协作机器人的方法可能会对改进的辅助和治疗设备和环境产生广泛影响，这些设备和环境可以降低老年人跌倒的风险；有助于缓解儿童的发育障碍；并帮助工人、士兵和急救人员完成身体任务。最终，这些定制方法可以提高鞋子和运动器材等日常物品的性能。该项目的科学目标是找到优化方法，有力地改善与用户的协作机器人互动，促进物理协作。向上扩展以支持各种设备、关节、行为、任务、环境和用户也是一个关键目标。该项目将探索一组优化的参数保持有效的时间，如何在不降低任务性能或设备接受度的情况下优化实际使用期间的设备行为，以及如何帮助用户确定何时以及如何参与行为的各种行为。就范围而言，该项目最初将侧重于机器人行为的参数化、优化算法的选择以及通过与协作机器人交互来增强人类学习能力，这些都可以在实现这一目标方面发挥重要作用。在方法方面，该项目将使用基于实验室的外骨骼模拟器以及便携式外骨骼设备，这些设备可以在户外开发和测试想法和方法，以及将提高对肌肉控制方式的理解的心理物理学研究。协同优化涉及协作机器人和用户同时优化他们的接口策略，从而为彼此的优化提供时变的目标。该项目将在物理协同机器人的背景下提高对协同优化的理论理解。该项目还将在现有的肌肉模型中添加新的分子现象，为人类与机器人的物理交互提供生理学基础和理解。这项工作的智力意义包括更好地理解人类是如何工作的，如何最有效地帮助人类，以及如何最好地为个人定制帮助。一个长期目标是为执行各种任务的个人构建一个定制交互策略库，并在用户执行所需任务时在线调优该库。

This project explores new ways to meet the needs of individual patients and users with customized active artificial limbs, motorized ankle and knee supports, and other forms of assistive robots. The effectiveness of such devices depends on accommodating person-to-person differences, such as in body types, walking gaits, and impairments. A preliminary study investigated the degree to which an individually adapted sequence of control patterns for a powered ankle exoskeleton -- modified as the subject used the device -- could reduce the metabolic cost of walking. This outcome showed that optimizing the assistive control pattern separately for each subject greatly increased the average benefits compared to the non-optimized device, and outperformed any previously documented approach. This project builds upon those preliminary results to find better and more robust customization methods for a wide range of assistive technology in diverse circumstances. This approach to "human-in-the-loop" optimization of human-collaborative robots could have a wide impact on improved assistive and therapeutic devices and environments that reduce the risk of falls in older adults; help mitigate developmental disorders in children; and assist workers, soldiers, and first responders with physical tasks. Ultimately these customization methods could improve performance of everyday items like shoes and exercise equipment.The scientific goal of the project is to find optimization approaches to robustly improve co-robotic interaction with users, facilitating physical collaboration. Scaling up to support a variety of devices, joints, behaviors, tasks, environments, and users is also a key goal. The project will explore how long a set of optimized parameters stay valid, how to optimize device behavior during actual use without reducing task performance or device acceptance, and how to assist a variety of behaviors where the user determines when and how a behavior is engaged. In terms of scope, the project will initially focus on how the parameterization of robot behavior, the choice of optimization algorithm, and enhancement of human learning by interaction with a co-robot can all play an important role in achieving this goal. In terms of methods, the project will use both laboratory-based exoskeleton emulators as well as portable exoskeleton devices that can go outdoors to develop and test ideas and approaches, as well as psychophysical studies that will improve understanding of how muscle is controlled. Co-optimization involves both the co-robot and the user optimizing their interface policies simultaneously, and thus presenting time-varying targets for each other's optimization. The project will improve theoretical understanding of co-optimization in the context of physical co-robotics. The project will also add new molecular phenomenon to current muscle models, to provide a physiological basis and understanding of human-robot physical interaction. The intellectual significance of this work includes better understanding of how humans work, how to most effectively assist humans, and how to best customize assistance for an individual. A long term goal is to build a library of customized interaction policies for an individual performing a variety of tasks, and tune the library online as the user does desired tasks.