Semi-autonomous sensory-motor control for upper limb prostheses

在过去的几十年里，人们努力在假肢和人类神经系统之间建立人机界面[1]。这些界面提供了一定程度的控制，在极少数情况下，还有可能向用户传递一些感觉。尽管如此，先进的双向感觉-运动控制接口的临床翻译一直是有限的。理想的假肢装置应该模仿生物假体的感知和驱动能力。考虑到人体四肢的复杂结构和神经控制，这是一个极端的挑战。虽然机电一体化技术近年来发展迅速，使我们能够接近这一理想，但通过最先进的人机界面提供的假肢和用户之间的信息传递仍然非常低。即使是先进的接口技术，神经接口也不太可能复制自然感觉-运动控制的所有方面。虽然假肢不能完全从神经上重新连接到使用者，以利用他们的决策和感觉/控制能力，但可以用它自己的智能来增强装置。这种方法是对辅助系统的一种与当前最先进水平完全不同的概念化，将机器人控制器从用户意图的简单解码器提升为与用户协作的智能代理，而不是完成功能任务。在这种方法中，用户只发送关于高级目标的信息(例如，抓取杯子)，而动作的细节(例如，具体的接近轨迹、抓取类型和大小)大多是在潜意识中展开的，从而模仿健康的受试者如何控制他们的肢体。因此，感觉-运动控制将变得半自主，就像大多数自然任务一样，在没有意识地控制所有身体自由度和整合感觉反馈的情况下执行。半自主控制假肢的方法已被初步探索，主要与感觉-运动回路的运动部分有关(例如，来自我们研究小组最近的论文[2])，而半自主的全感觉-运动回路在假肢系统中还被探索过。在这个项目中，我们建议开发一种用于假肢的半自主控制系统，该系统在用户级(意识感觉-运动控制)和人工智能代理级别(自主感觉-运动控制)在闭环系统中集成运动命令和感觉信息。这两个感觉-运动环共同承担假体最终激活的任务。

In the past decades, efforts have been made to establish man-machine interfacesbetween prostheses and the human nervous system [1]. These interfaces have providedsome level of control and, in rare cases, the possibility to transmit some sensation to theusers. Nonetheless, clinical translation of advanced bidirectional sensory-motor controlinterfaces has been limited. An ideal prosthetic device should mimic the sensing and actuation capabilities of thebiological counterpart. Considering the sophisticated structure and neural control ofhuman limbs, this is an extreme challenge. While the mechatronic technology has beendeveloping rapidly in recent years, allowing us to come close to this ideal, the informationtransfer between the prosthesis and the user offered by state-of-the-art man-machineinterfaces is still extremely low. Even advancing the interfacing technology, it is veryunlikely that neural interfaces can replicate all aspects of natural sensory-motor control. While the prosthesis cannot be fully neurally "re-connected" to the user to exploit theirdecision-making and sensory/control abilities, it is possible to enhance the device with itsown intelligence. This approach is a radically different conceptualization of an assistivesystem with respect to current state-of-the-art, promoting the robotic controller from asimple decoder of the user intent into an intelligent agent collaborating with the user inaccomplishing functional tasks. Within this approach, the user transmits information onlyabout a high-level goal (e.g., grasping a cup) while the details of the movements (e.g.,specific approach trajectory, grasp type and size) unfold mostly subconsciously, mimicking thereby how healthy subjects control their limbs. The sensory-motor control wouldtherefore become semi-autonomous, as for most natural tasks that are executed withoutconsciously controlling all the body degrees of freedom and integrating sensory feedback.The semi-autonomous approach to prosthesis control has been preliminary exploredmainly in relation to the motor part of the sensory-motor loop (e.g., from our researchgroup see the recent paper [2]) while semi-autonomous full sensory-motor loops havenever been explored in prosthetic systems. In this project we propose the development of a semi-autonomous control system forprostheses that integrate motor commands with sensory information in closed-loopsystems both at the user level (conscious sensory-motor control) and at an AI agent level(autonomous sensory-motor control). These two sensory-motor loops share the task ofthe ultimate activation of the prosthesis.