NRI: Biologically Inspired Feedback Control of Robots Interacting with Humans to Cooperate and Assist with Repetitive Movement Tasks

生物启发的反馈控制机器人与人类互动，以合作和协助重复的运动tasksThe项目将解决如何控制机器人的运动，使它可以与人类合作，以协助他们在重复的任务的基本问题。人体的摆动运动是人类完成各种任务的基本手段。这种重复性运动包括基本的生活功能，如心跳、呼吸、进食（咀嚼）、行走;基本的日常任务，如刷牙、洗脸;家务杂务，如擦窗户、扫地;健康/娱乐活动，如跳舞、游泳、骑自行车、划船;以及制造劳动，如在工厂装配线上移动物体。在许多情况下，机器人和机械装置可以帮助人类进行这种移动。机器人操纵器和人的手臂可以抓住共同的工具以在重复性任务上一起工作，其中前者通过提供力和稳定性来帮助后者以减少人的负担。可穿戴外骨骼以补充老年人及具有神经病症或身体残疾的患者的能力降低或为老年人及具有神经病症或身体残疾的患者提供康复。因此，设计良好的振荡运动辅助装置将大大有助于提高人类生活质量。设计这种辅助设备的机器人机构无疑是一项具有挑战性的任务。同样具有挑战性的是控制算法的设计，命令致动器和管理机器人设备的运动。最先进的控制技术允许设计人员对机器人进行编程，以实现速度，精度和鲁棒性的规定运动，例如在工业机械手中。然而，如果这样的机器人与人类互动，它们会被认为是僵硬的，顽固的，甚至是危险的，因此不适合作为直接支持人类的合作机器人。我们需要的是控制算法，使机器人理解人类的意图，与人类合作，而不坚持他们的预编程操作，并协助人类的任务。本课题的研究重点是开发这样的算法。本基础研究的目的是，为了稳定人的振动，并通过提供辅助力来减轻人的负担，建立与人互动的一般机器人系统的反馈控制器的系统化设计方法。控制架构的灵感来自中央模式发生器（CPG）-神经元电路，命令肌肉收缩，以实现动物运动过程中有节奏的身体运动。CPG是有吸引力的工程应用，由于其能够符合其振荡的自然动态变化的环境，通过感官反馈。这项探索性研究将调查CPG架构的潜力，为实现协助人类执行振荡任务的合作机器人的新系统设计提供可行的基础。控制器实现为相同单元的互连，模拟神经元动力学。该问题被制定为搜索的互连，使机器人-人-CPG系统有一个稳定的极限环，其中人类决定一个适当的振荡和CPG控制的机器人协助。将采用多变量谐波平衡的方法来获得满足振荡规范的可行互连矩阵的凸表征。设计方法的近似性质将通过广泛的模拟以及机器人机械手的物理实验来补充。虽然控制理论的中心主题是围绕动力系统的平衡点进行调节，但产生协调自主振荡的能力在许多工程应用中非常有用。这里提出的基础研究将提供一个初步的垫脚石合作模式生成的反馈控制的新范式。

Biologically inspired feedback control of robots interacting with humans to cooperate and assist with repetitive movement tasksThe project will address the fundamental problem of how to control the motion of a robot so that it can cooperatively work with humans to assist them in repetitive tasks. Oscillatory body movements constitute an elementary means for various tasks in human living. Such repetitive movements include essential life functions such as heart beat, breathing, eating (chewing), walking; basic daily tasks such as brushing teeth, washing face; house-hold chores such as cleaning windows, sweeping floor; health/entertainment activities such as dancing, swimming, cycling, rowing; and manufacturing labors such as moving objects in factory assembly lines. Robots and mechanical devices that assist such human movements would be found useful in a number of contexts. A robotic manipulator and a human arm may grab a common tool to work together on repetitive tasks where the former assists the latter by providing force and stability to reduce burden on the human. An exoskeleton may be worn to complement reduced capability of, or provide rehabilitations for, elderly people and patients with neurological disorders or physical disabilities. Thus, well-designed assistive devices for oscillatory movements would significantly contribute to improving quality of human life. Design of robotic mechanisms for such assistive devices is surely a challenging task. Equally challenging is the design of control algorithms that command the actuators and govern the motion of the robotic device. The state-of-the-art control technologies allow a designer to program a robot to achieve prescribed motion with speed, precision, and robustness, as seen for instance in industrial manipulators. However, if such robots interact with humans, they would be perceived as stiff, stubborn, or even dangerous, and are therefore not suitable as co-robots in direct support of humans. What is needed is control algorithms that make robots understand human intentions, cooperate with humans without insisting on their preprogramed operations, and assist with human tasks. Development of such algorithms will be the focus of this project.This basic research aims to establish a systematic method for designing a feedback controller for a general robotic system interacting with a human to stabilize the oscillation intended by the human and to reduce the burden on the human by providing assistive forces. The control architecture is inspired by the central pattern generator (CPG) -- neuronal circuits that command muscle contractions to achieve rhythmic body movements during animal locomotion. CPGs are attractive for engineering applications due to its ability to conform their oscillations to natural dynamics of a varying environment through sensory feedback. This exploratory research will investigate the potential of the CPG architecture to provide a viable foundation for a new system design for achieving co-robots that assist humans to execute oscillation tasks. The controller is realized as an interconnection of identical units, emulating neuronal dynamics. The problem is formulated as the search for the interconnection such that the robot-human-CPG system has a stable limit cycle in which human decides an appropriate oscillation and CPG-controlled robot assists. The method of multivariable harmonic balance will be employed to obtain a convex characterization of feasible interconnection matrices that meet oscillation specifications. The approximate nature of the design method will be complemented by extensive simulations as well as physical experiments on robotic manipulators. While the central theme of control theory has been the regulation around an equilibrium point of a dynamical system, capability of generating coordinated autonomous oscillations can be extremely useful in many engineering applications. The basic research proposed here will provide an initial stepping stone toward a new paradigm for cooperative pattern generations by feedback control.