SBIR Phase I: Personalized Wearable Device that Learns, Adapts to Users, and Provides Enhanced Metrics and Assessments

这个小型企业创新研究（SBIR）第一阶段项目的更广泛的影响/商业潜力在于解决消费者广泛采用可穿戴技术设备用于智能健康用例的长期障碍。 主要的障碍是，当前的设备在将身体佩戴的传感器数据转换为活动/手势数据时非常不准确。第二个障碍是，没有可靠的手势数据，就没有向用户提供有意义的反馈的机制。 为了解决这些障碍，我们提出了一种创新的方法，从身体穿戴传感器数据的手势/活动识别。该软件可以快速学习和适应用户的运动特质，在一个计算量小但功能强大的系统。商业化计划是将该软件授权给设备制造商，让他们将其嵌入到运动传感产品中。 以前无法分析的产品和用例现在可以启用跟踪，从而为最终用户带来价值，并为制造商扩大经济机会。 拟议中的软件将使个人能够管理他们的健身和康复方案;运动教练和医生将拥有前所未有的工具来监测门诊治疗;该领域将拥有大量的行为数据，以挖掘研究意义和其他社会成果。 拟议的项目将导致软件可以与可穿戴技术设备集成，嵌入到硅或作为一个独立的移动的应用程序。该软件将能够1）准确地识别多个活动/手势，尽管个体差异和其他噪声因素，2）提供这些活动的可靠和有效的属性（指标），特别是运动范围，路径效率和功率，以及3）为用户反馈和研究目的提供有意义的数据。尽管现在可用的相关设备和技术过多，但是当前技术缺乏关于所识别的手势的数量、识别的准确性、对设备上学习的支持以及提供有效且可靠的手势度量的能力。 所提出的技术将解决当前功能、准确性和鲁棒性方面的限制和市场需求。 总体的技术问题是，软件是否可以被开发为充分补偿传感器漂移和运动数据中的其他类型的不确定性，在日益复杂的物理行为和输出（度量）要求所带来的复杂程度。 预期的技术成果包括新的算法，基本的性能界限，软件原型，测试和验证研究。第一阶段的示范如果成功，将导致这些功能和其他功能的全面软件开发。

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase I project lies in addressing a long-standing roadblock to widespread consumer adoption of wearable tech devices for Smart Health use-cases. The primary barrier is that the current devices are woefully inaccurate when it comes to converting body-worn sensor data into activity/gesture data. A second barrier is that without reliable gesture data, there is no mechanism for providing meaningful feedback to users. To address these barriers, we propose an innovative approach to gesture/activity recognition from body-worn sensor data. The software can rapidly learn and adapt to user's motion idiosyncrasies, in a computationally lightweight but powerful system. The commercialization plan is to license the software to equipment manufacturers for them to embed it in their motion sensing products. Products and use-cases that have been impossible to analyze previously can now have tracking enabled, thus driving value for the end-user and expanding economic opportunities for the manufacturers. The proposed software will empower individuals to manage their fitness and rehabilitation regimens; sports trainers and physicians will have unprecedented tools for monitoring outpatient treatment; and the field will have troves of behavior data to mine for research implications and additional societal outcomes. The proposed project will lead to software that can be integrated with wearable tech devices, either embedded into silicon or as a stand-alone mobile app. The software will be capable of 1) accurately recognizing multiple activities/gestures despite individual variances and other noise factors, 2) providing reliable and valid attributes (metrics) of those activities, specifically range of motion, path efficiency, and power, and 3) providing meaningful data for user feedback and research purposes. Despite the plethora of related devices and technologies now available, current technology is lacking as to the number of gestures identified, accuracy of identification, support for on-device learning, and the ability to provide valid and reliable gesture metrics. The proposed technology will address current limitations in and market demand for functionality, accuracy, and robustness. The overall technical question is whether the software can be developed to adequately compensate for sensor drift and for other types of uncertainties in motion data, at the level of complexity posed by increasingly sophisticated physical behaviors and output (metric) requirements. The anticipated technical results include novel algorithms, fundamental performance bounds, software prototypes, and testing and validation studies. Phase I demonstrations, if successful, will lead to full-scale software development of these and additional functions.