Brain-Robot Interface: A Robust, High Performance Predictive Control Algorithm

描述（由申请人提供）：项目摘要：拟议研究的目的是开发一种改进的“思想引导”机器人控制算法，使用人脑的脑电图（EEG），用于机器人设备执行的某些动作。我认为，正在开发的现有脑机接口（BMI）的性能有限的原因之一，是由于大脑信号的动力学对于现有的控制算法来说太复杂而无法有效。为了克服这一点，我设想在神经元动态模型识别和控制算法的发展，这将有助于更好地了解复杂的神经元系统，并提供更好的性能控制系统的范式转变。在这个项目中，我计划研究假设：1）基于模型的预测方法和最优控制策略将大大提高BMI中的信号提取和控制性能，超过现有的非基于模型和非最优方法。我在这里提出一个两部分的方法：1）响应于音频/视觉提示来获取非侵入性人类头皮EEG，并且使用现有的和提出的基于模型的技术来比较和对比EEG特征，以便开发状态空间格式的数据驱动的经验模型，以及2）设计MFC算法，与常用的滤波和/或比例反馈控制算法不同，将预测机器人设备在未来的一段时间内的期望移动，称为预测范围，将使用优化算法来计算当前采样时间处的控制移动，并且将具有使用多个控制器参数在线调谐的能力，以便有效地控制机器人设备。这个K25指导职业奖的更广泛的目标是将候选人的控制工程背景应用于神经应用，并在这两个研究领域之间建立一个接口。除了在宾夕法尼亚州立大学工程科学和力学系开展拟议的研究外，拟议的职业发展计划还包括通过课程、研讨会和其他教学手段进行广泛的培训，这将使候选人为神经问题解决方案的独立学术生涯奠定坚实的基础。 与公共卫生的相关性：在拟议的研究中开发的技术将通过将现代控制工程与神经生物学相结合来促进对复杂神经生物学行为的理解。这意味着，通过开发更智能、更有效的神经修复术，患有神经系统疾病（如脑或脊髓损伤）的人将得到更好的帮助。

DESCRIPTION (provided by applicant): Project Summary: The objective of the proposed research is to develop an improved "thought-guided" robotic control algorithm using electroencephalography (EEG) from the human brain for certain actions to be performed by a robotic device. One of the reasons of limited performance of existing brain machine interfaces (BMI) being developed to assist people with neurological disorders to conduct motor activities, I believe, is due to the fact that the dynamics of brain signals are too complex for the existing control algorithms to be efficient. To overcome this, I envision a paradigm shift in the neuronal dynamic model identification and control algorithm development that will help better understand the complex neuronal systems and provide improved performance to the control system. In the proposed project, I plan to investigate the hypothesis: 1) Model-based predictive approach and optimal control strategies will improve signal extraction and control performance in BMI substantially over existing non-model based and non-optimal approaches. I here propose a two part approach: 1) acquire non-invasive human scalp EEG in response to audio/visual cues, and compare and contrast EEG features using existing and proposed model-based techniques in order to develop a data driven, empirical model in state space format, and 2) design an MFC algorithm which, unlike the commonly used filtering and/or proportional feedback control algorithms, will predict the desired movement of the robotic device over a period of time in the future called the prediction horizon, will use an optimization algorithm to calculate the control move at the current sampling time, and will have the ability to be tuned online using a number of controller parameters in order to efficiently control the robotic device. The broader aim of this K25 mentored career award is to apply the candidate's control engineering background to neurological applications and develop an interface between these two research areas. The proposed career development plan includes, in addition to carrying out the proposed research in the Engineering Science and Mechanics Department at Penn State University, extensive training through courses, workshops, and other didactic means which will allow the candidate to build a strong foundation for an independent academic career in neurological problem solutions. Relevance to Public Health: Technology developed in the proposed research will advance the understanding of the complex neurobiological behavior by merging modern control engineering with neurobiology. It has implications that people with neurological disorders such as brain or spinal cord injuries will be better assisted through the development of smarter, more effective neural prosthetics.