Development of a Novel EMG-Based Neural Interface for Control of Transradial Prostheses with Gripping Assistance

项目摘要 上肢截肢会使许多基本任务变得困难或几乎不可能。近年来，研究 可以从残肢的肌电图（EMG）信号预测运动意图的算法已经导致 假手的发展，允许控制多个自由度（DOF），并已恢复 上肢的基本功能一些最先进的商用假手 使用基于机器学习的控制方案，称为EMG模式识别（PR）。许多EMG PR 方法预测运动类别（例如，手打开/闭合）并将电机的速度设置为与运动类别成比例。 EMG信号的强度。一些新提出的方法涉及同时控制 多自由度然而，所有这些控制方案都允许用户对施加到物体上的力进行最小限度的控制 这使得抓握和运输易碎物品变得困难。 该项目的总体目标是开发一种新的控制方案，允许同时控制 经桡动脉假体的多个DOF的位置以及当物体已经制造完成时对抓握力的控制 与假肢的指尖接触。为实现这一目标，本提案包括以下2项内容： 目的：1）开发一种新的共享控制框架，用于实时上肢假肢控制和抓取， 2)评估共享控制框架的性能和认知工作负荷。共享控制器将 使用人工神经网络（ANN）将肌电信号的特征映射到关节扭矩和前向动力学 模型来计算关节运动学。将收集受试者的EMG和关节运动数据，并进行强化 学习算法将用于训练人工神经网络，以最小化估计和测量之间的误差 岗位一个附着在假手指尖的力传感器将检测何时与物体接触 并测量握力。掌指（MCP）关节的估计扭矩将 并且PID控制器将驱动测量的抓握力到该期望的抓握力 握力为了评估该框架，将使用虚拟任务来测试受试者控制握力的能力 通过让它们跟随显示器上显示的给定力轨迹来控制手的运动。然后，受试者将同时使用 共享的控制器和EMG PR完成2个功能任务，包括运输易碎/变形 对象任务将在有和没有精神上要求很高的双重任务的情况下完成， 性能将被用来估计认知负荷。 这项拟议的工作预计将介绍一种控制经桡动脉假体的方法， 基于位置的多自由度控制和精确控制假手施加到物体上的抓握力 不同程度的服从。这种方法可以降低物体抓取的难度和心理要求 任务，并导致更高的接受率的动力上肢假肢。

PROJECT SUMMARY An upper limb amputation can make many basic tasks difficult or nearly impossible. In recent years, research in algorithms that can predict motion intentions from electromyographic (EMG) signals of a residual limb has led to the development of prosthetic hands that allow control of multiple degrees of freedom (DOF) and has restored the basic functionality of an upper-limb. Some of the most advanced commercially available prosthetic hands use a machine learning-based control scheme known as EMG pattern recognition (PR). Many EMG PR approaches predict a motion class (e.g. hand open/close) and set the velocity of the motors proportional to the magnitude of EMG signals. Some new proposed approaches involve simultaneously controlling the position of multiple DOF. However, all of these control schemes allow users minimal control of the force applied to objects grasped by the prosthetic hand which makes holding and transporting fragile objects difficult. The overall objective of this project is to develop a novel control scheme that allows simultaneous control of the positions of multiple DOF of a transradial prosthesis as well as control of grip force when an object has made contact with the fingertips of the prosthesis. To achieve this objective, this proposal consists of the following 2 aims: 1) Develop a novel shared control framework for real-time upper limb prosthesis control and gripping and 2) Evaluate the performance and cognitive workload of the shared control framework. The shared controller will use an artificial neural network (ANN) to map the features of EMG signals to joint torque and a forward dynamics model to calculate joint kinematics. EMG and joint motion data will be collected from subjects and a reinforcement learning algorithm will be used to train the ANN to minimize the error between estimated and measured joint positions. A force sensor attached to the fingertip of a prosthetic hand will detect when contact with an object has been made and measure the grip force. The estimated torque of the metacarpophalangeal (MCP) joint will be used to estimate a desired grip force and a PID controller will drive the measured grip force to this desired grip force. To evaluate the framework, a virtual task will be used to test a subjects’ ability to control the grip force of the hand by having them follow a given force trajectory displayed on a monitor. Then, subjects will use both the shared controller and EMG PR to complete 2 functional tasks involving transporting fragile/deformable objects. Tasks will be completed with and without a mentally demanding dual task and the differences in performance will be used to estimate cognitive loads. This proposed work is expected to introduce a method of controlling transradial prostheses that provides reliable position-based control of multiple DOF and precise control of the grip force the prosthetic hand applies to objects with various levels of compliance. This method can reduce the difficulties and mental demands of object grasping tasks and lead to a higher acceptance rate of powered upper limb prostheses.