Neuromorphic encoding of tactile stimuli to provide naturalistic sensory feedback in upper limb prostheses

项目摘要 该研究项目的动机是通过以下方式改善截肢者的生活 触觉刺激的自然感觉反馈。今天，假肢依靠解码用户的意图， 神经或肌电图（EMG）信号的测量。这些复杂的 机器人设备不能在没有结合评估环境的传感器的情况下实现， 一种与用户无缝沟通的方式。神经假体可以实现这种无缝的交流 通过直接与截肢者的神经系统连接并刺激神经， 与假体和环境之间的相互作用相对应的感觉。这么做 自然地，来自并入假体中的传感器的模拟读数必须被编码到 神经系统的语言：尖峰活动的模式。我的目标是改善感官反馈， 通过探索如何将触觉传感器的信息转化为神经元样信息， （神经形态的）尖峰被用于刺激反馈。我将研究触觉是如何编码的 在生物学中，然后使用计算模型从现象学上重建信号处理链， 将用真实世界的纹理数据集进行测试。这些模型的输出将被分类以进行验证 并将纹理信息的成功编码量化为神经形态尖峰活动。纹理 由于其丰富的时空结构，可以作为开发这些模型的一个很好的测试案例。 具体目标1 -触觉刺激的神经形态编码和处理-我将使用Izhikevich 神经元模型，以重建SA和RA机械感受器响应于纹理刺激的尖峰活动 应用于触觉传感阵列。我会开发新的算法来改变尖峰模式 扫描速度和施加的力。这将导致速度和力不变的表示， 质地。 具体目标2 -触觉信息的神经形态压缩-最初，一个幼稚的通道 一种使用脉冲串距离来评估不同输入之间的互信息的选择算法 通道将压缩触觉信息，以选择最佳的一组感测通道来通过， 刺激.更高级的方案将联合收割机的输入组合在一起，以实现更有效的信息编码 并丰富最终输出尖峰模式的信息内容。人工纹理分类 将被用来评估这些方法的能力，有效地保留相关的纹理信息。 从根本上说，目标1侧重于触觉刺激的鲁棒表示， 探索性条件，而目标2侧重于这些刺激的有效表征。完成后， 这项工作将为通过外周神经向截肢者提供更自然的感觉反馈提供基础 在日常生活中使用假肢时，这种刺激将导致更好的功能结果。

PROJECT SUMMARY This research project is motivated by the goal of improving the lives of amputees through naturalistic sensory feedback of tactile stimuli. Today, prostheses rely on decoding user intention through measurement of neural or electromyographic (EMG) signals. The full potential of these sophisticated robotic devices cannot be realized without the incorporation of sensors that evaluate the environment and a way to seamlessly communicate with the user. Neural prostheses can enable this seamless communication by interfacing directly with the nervous system of amputees and stimulating the nerves in order to elicit sensations corresponding to the interaction between the prosthesis and the environment. To do this naturalistically, the analog readings from sensors incorporated into the prosthesis must be encoded into the language of the nervous system: patterns of spiking activity. My goal is to improve sensory feedback for amputees by exploring how information from tactile sensors can be transformed into neuron-like (neuromorphic) spikes to be used for stimulation feedback. I will examine how tactile sensing is encoded in biology and then phenomenologically recreate the signal processing chain using a computational model that will be tested with a real-world texture dataset. The output of these models will be classified to verify and quantify the successful encoding of texture information as neuromorphic spiking activity. Texture serves as a good test case to develop these models because of its rich spatiotemporal structure. Specific Aim 1 – Neuromorphic Encoding and Processing of Tactile Stimuli – I will use the Izhikevich neuron model to recreate the spiking activity of SA and RA mechanoreceptors in response to texture stimuli applied to a tactile sensing array. I will develop new algorithms to transform the spiking patterns to account for scanning speed and applied force. This will result in a speed- and force-invariant representation of texture. Specific Aim 2 – Neuromorphic Compression of Tactile Information – Initially, a naïve channel selection algorithm that uses spike train distance to evaluate mutual information between different input channels will compress tactile information to select an optimal set of sensing channels to pass through to stimulation. A more advanced scheme will combine inputs together for more efficient information encoding and to enrich the information content of the final output spiking patterns. Artificial texture classification will be used to evaluate the capability of these methods to efficiently retain relevant texture information. Fundamentally, Aim 1 focuses on robust representations of tactile stimuli independent of exploratory conditions, while Aim 2 focuses on efficient representations of those stimuli. When completed, this work will provide the basis for more naturalistic sensory feedback to amputees through peripheral nerve stimulation which will result in better functional outcomes when using prostheses in their daily lives.