Motor Learning for the Control of an Assistive Device

这些研究的目标是了解如何利用和训练身体的运动来控制电动轮椅。先进的轮椅技术经常被大量潜在的轮椅使用者视为障碍。提出了一种消除这一障碍的方法，其基础是使辅助技术适应患者剩余的无限制活动和加强运动学习。这一探索性项目旨在确定这种方法的可行性，并开发基于识别自然运动和使用虚拟现实(VR)的训练方法。拟议的研究将在四肢瘫痪脊髓损伤患者中进行，这些患者有完整或不完全的颈椎损伤。健康的志愿者也将参与这些研究，以微调实验仪器，并提供评估学习和协调能力的参考基线。受试者将穿上一件新颖的上半身感应服装。这件衣服产生的总共52个电信号将受到手腕、肘部、肩膀和躯干运动的调制。这些信号将被映射到模拟轮椅的速度命令中。受试者将佩戴VR护目镜和头部追踪器，这将使他们从模拟轮椅的角度身临其境地观看计算机生成的环境。虚拟现实环境和可穿戴信号技术的结合将为评估培训协议提供一个框架，而这在实际的轮椅上是不可行的。拟议的研究有两个具体目标：(目标1)。为了识别用于通过不受限制的上半身运动来控制虚拟轮椅的运动基元，将使用和比较三种成熟的信号处理技术-主成分分析、独立成分分析和ISOMAP-以从服装信号中提取低维信号模式(目标2)。识别促进运动学习的地图和程序。从目标1中提取的信号模式将用于设计和测试从受试者运动到轮椅命令的新转换。一种著名的机器学习技术-最小均方梯度下降-将进行测试，以将受试者的自然运动基元与一组适当的控制信号匹配到轮椅。最后，安全的虚拟现实环境将允许我们测试通过逐渐加快轮椅运动或逐渐减慢轮椅运动来学习是最有效的。这些研究结果有望指导基于人类运动学习和自适应控制工程的辅助装置新技术的开发。许多残疾人在利用辅助技术方面面临着困难的挑战。特别是，由于患者需要学习操作他们的控制设备，电动轮椅的安全和有效使用受到限制。拟议的研究将探讨扭转这种情况的可能性，并利用先进技术使控制设备适应患者的剩余技能。

The goal of these studies is to understand how movements of the body can be harnessed and trained to control electrically powered wheelchairs. Advanced wheelchair technology is often perceived to be a barrier by a large number of potential wheelchair users. An approach is proposed for the removal of this barrier based on adapting the assistive technology to the residual unconstrained mobility of the patients and on enhancing motor learning. This exploratory project aims at establishing the feasibility of such an approach and at developing training methods based on the identification of natural motions and on the use of virtual reality (VR). The proposed studies will be carried out on quadriplegic spinal cord injured patients with complete or incomplete cervical injuries. Healthy volunteers will also participate in these study to fine-tune the experimental apparatus and to provide a reference baseline to assess learning and coordination. Subjects will wear a novel upper-body sensing garment. A total of 52 electrical signals generated by the garment will be modulated by movements of the wrist, elbow, shoulder and torso. These signals will be mapped into the velocity commands for a simulated wheelchair. Subjects will wear VR-goggles and a head tracker, which will provide them with a immersive view of a computer-generated environment from the perspective of the simulated wheelchair. The combination of virtual reality environments and wearable signal technology will provide a framework for evaluating training protocols that would not be feasible with actual wheelchairs. The proposed studies are organized in two specific aims: (Aim 1.) To identify motor primitives for the control of a virtual wheelchair by unrestricted upper body motions Three well-established signal processing techniques - Principal Component Analysis, Independent Component Analysis and Isomap - will be used and compared for extracting low-dimensional signal patterns from the garment signals (Aim 2.) To identify maps and procedures that facilitate motor learning. The signal patterns extracted from Aim 1 will be used to design and test new transformations from subject motions to wheelchair commands. A well- known machine learning technique -least mean squares gradient descent - will be tested for matching the natural motor primitives of the subjects with an appropriate set of control signals to the wheelchair. Finally the safe VR environment will allow us to test whether it is most efficient to learn by gradually speeding up wheelchair motions or by gradually slowing them down. The results of these studies are expected to guide the development of new technology for assistive devices based on human motor learning and on engineering of adaptive control. Many disabled individuals are facing difficult challenges to take advantage of assistive technologies. In particular the safe and efficient use of powered wheelchair is limited by the need for patients to learn to operate their control apparatus. The proposed studies will investigate the possibility to reverse this situation and take advantage of advanced technologies for adapting the control apparatus to the residual skills of the patients.