CAREER: Adaptive Sonification to Improve Balance during Everyday Mobility

这项教师早期职业发展（Career）资助将使用机器学习和可穿戴技术来识别容易跌倒的老年人的平衡和步态缺陷，并提供个性化的听觉生物反馈（“声波生物反馈”），旨在改善行走时的动态平衡。转向与老年人反复跌倒有关，因为它在平衡和改变方向之间施加了机械冲突。老年人摔倒经常导致受伤，有时甚至死亡。根据一些统计，在任何一天中，转身走路的步数可能占到总步数的50%。再训练在转身时使用的平衡策略有可能减少跌倒的风险。该项目将确定转弯时使用的个人特定平衡策略与其他因素之间的关系，包括生理和认知能力（如力量和空间推理能力）以及环境因素（如是否存在障碍物）。这些信息将用于设计个性化的听觉生物反馈，以传递有关运动运动学和足部力量的信息，从而促进行走过程中的动态平衡。最初的一组实验将测试人类对动态平衡信息的声音生物反馈的感知和运动反应。另外的实验将测试容易摔倒的老年人使用个性化生物反馈来改善转弯时动态平衡的能力。该项目通过开发一种机器学习方法来诊断老年人的平衡缺陷，以及一种新的超声生物反馈方法来改善老年人的动态平衡，从而促进了国民健康。该项目包括一项教育和推广计划，其中包括一项创新的“驻地艺术家”计划，该计划将以一种吸引工程师和年轻艺术家的方式，让未被充分代表的群体接触到尖端的工程和科学研究。这个项目测试了一个假设，即交互式声波生物反馈可以改善有跌倒风险的老年人在转身走路时使用的平衡策略。这个项目有三个研究目标。首先，PI将进行人体实验，以表征转弯时使用的动态平衡策略、生理和认知能力以及环境因素之间的个人特定关系。参与者将在实验室和户外的可控环境中行走和转身，同时佩戴可以实时感知和传输运动运动学和地面反作用力的技术。对于第二个目标，PI将确定如何使用超声来训练特定的平衡策略，以及在不依赖于并发生物反馈的情况下保留和使用这些策略的程度。与声音设计师合作，PI将开发声音场景，在不受约束的身体运动中对平衡指标进行声音处理，评估他们在重复训练过程中实时调节稳定性的能力，并测试他们引发长期行为改变的能力。在第三个目标中，PI将测试机器学习模型在与临床医生诊断相比的回合中诊断个人特定平衡缺陷的能力，并设计个性化的生物反馈音景，以减轻临床医生诊断的平衡缺陷。在这里，PI将使用因子条件限制玻尔兹曼机器生成个人特定的、回合类型特定的和缺陷特定的模型，这些模型将能够生成个性化的声音生物反馈，以改善回合期间的平衡。这个项目将推进对智能机器的基本理解，智能机器可以通过物理交互来传达意图或塑造人类用户的行为。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

This Faculty Early Career Development (CAREER) grant will use machine learning and wearable technology to identify balance and gait deficits in fall-prone older adults and to deliver personalized auditory biofeedback ("sonified biofeedback") designed to improve dynamic balance while walking. Turning has been linked to recurrent falls in older adults because it imposes mechanical conflict between balancing and changing direction. Falls in older adults frequently lead to injury and sometimes death. By some accounts, turning-while-walking can comprise up to 50 percent of steps taken in any given day. Retraining balance strategies used during turning-while-walking has potential to reduce fall risk. This project will identify relationships between person-specific balance strategies used during turning and other factors including physiological and cognitive capabilities - such as strength and capacity for spatial reasoning - as well as environmental factors such as the presence or absence of obstacles. This information will be used to design personalized auditory biofeedback tuned to convey information about movement kinematics and foot forces in a way that facilitates dynamic balance during walking. An initial set of experiments will test human perception and motor responses to sonified biofeedback of dynamic balance information. Additional experiments will test the ability of fall-prone older adults to use personalized biofeedback to improve dynamic balance during turns. This project advances the national health by developing a machine learning approach to the diagnosis of balance deficits in older adults as well as a novel sonified biofeedback approach to improving dynamic balance in that population. The project includes an education and outreach plan, including an innovative “Artist in Residence” program, that will expose underrepresented groups to cutting edge engineering and scientific research in a way that is engaging for engineers and young artists alike. This project tests the hypothesis that interactive sonified biofeedback can improve balance strategies used by fall-risk older adults during turning-while-walking. This project has three research objectives. In the first, the PI will conduct human subjects experiments to characterize person-specific relationships between dynamic balance strategies used during turns, physiological and cognitive capabilities, and environmental factors. Participants will walk and turn within the controlled setting of the research lab and outdoors while wearing technology that can sense and transmit movement kinematics and ground reaction forces in real-time. For the second objective, the PI will establish how sonification can be used to train specific balance strategies and the extent to which those strategies can be retained and used without dependency on concurrent biofeedback. Working in conjunction with a sound designer, the PI will develop soundscapes that sonify balance metrics during unconstrained body movements, evaluate their ability to modulate measures of stability in real-time during repeated training sessions, and test their ability to elicit long-lasting behavioral change. In the third objective, the PI will test the ability of machine learning models to diagnose person-specific balance deficits during turns compared to clinician diagnoses, and to design personalized biofeedback soundscapes that can mitigate clinician-diagnosed balance deficits. Here, the PI will use Factored Conditional Restricted Boltzman Machines to generate person-specific, turn-type specific, and deficit-specific models that will be able to generate personalized sonified biofeedback to improve balance during turns. This project will advance fundamental understanding of an intelligent machine can communicate intent or otherwise shape the behavior of its human user through physical interaction.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.