CPS: Synergy: Collaborative Research: Learning control sharing strategies for assistive cyber-physical systems

辅助网络物理系统的学习控制共享策略辅助机器，如电动轮椅，肌电假肢和机械臂，可以提高严重运动障碍患者的独立性和能力。随着这些辅助网络物理系统（cps）的进步，更灵巧、更有能力的机器有望彻底改变运动障碍患者在社会中与亲人互动的方式，并独立照顾自己。然而，随着这些机器变得越来越强大，它们也变得越来越复杂。这就提出了一个问题：如何控制这种增加的复杂性？本文提出了一种新的范例，用于控制复杂的辅助网络物理系统（cps），如安装在轮椅上的机械臂，通过简单的低维控制接口，如二维操纵杆或一维Sip-N-Puff接口，使严重运动障碍的人可以使用。传统的接口只覆盖了一部分控制空间，在远程操作时，需要在不同的控制模式之间切换以访问整个控制空间。机器人自动化可以用来预测何时在不同的控制模式之间切换。该方法与辅助领域的大多数控制共享方法不同，这些方法要么划分控制空间并将不同的部分分配给机器人和人类，要么增加人类的控制信号以弥合维度差距。如何在辅助域内最好地共享控制仍然是一个悬而未决的问题，这种方法的一个吸引人的特点是，由于用户的信号没有改变或增强，因此用户可以最大限度地保持控制。通过提高严重运动障碍和/或瘫痪患者的独立性，对公共卫生产生重大影响。多方努力将促进我们的成果的大规模部署，包括与辅助机械臂制造商Kinova的合作，以及与芝加哥康复研究所的合作。提出了一种基于混合动力系统理论的辅助模式切换形式，旨在减轻远程操作高维辅助机器人的负担。通过将该CPS建模为混合动力系统，可以将辅助建模为对期望成本函数的优化。系统对用户目标的不确定性可以通过部分可观察马尔可夫决策过程来建模。这个模型为学习用户偏好提供了自然的框架。通过用户研究，本项目旨在解决以下研究问题：（Q1）费用：模式切换的成本有多高？（Q2）定制需求：我们是否需要从特定用户那里学习模式切换？（Q3）学习辅助：我们如何从用户那里学习模式切换范例？（Q4）目标不确定性：在目标不确定性下，援助应如何行动？用户将如何回应？该提案利用了团队在操作、算法开发和部署现实世界机器人系统方面的共享专业知识。该提案还利用了团队在部署先进操作平台、机器人运动规划和操作、人机协同以及机器人从人类演示、控制政策适应和人类康复中学习方面的互补优势。这项工作的目标是让严重瘫痪的用户更容易操作机械臂。在远程操作过程中，需要控制多个自由度（DoF），从而产生模式切换。即使使用2轴和3轴操纵杆，切换本身也很麻烦，并且在更有限的（1-D）接口下变得令人望而却步。简化切换操作不仅减轻了那些已经能够操作机械臂的人的负担，而且可能向目前无法使用辅助机械臂的人群开放。这项工作显然是协同的：在机器人操作，人类康复，控制理论，机器学习，人机交互和临床研究的交叉点。该项目通过开发系统与用户之间交互动力学的新模型来解决CPS的科学问题，通过开发具有强大理论基础的新界面和交互模式来解决CPS的技术问题，并通过将我们的算法部署到真实的机器人硬件上以及对健全和脊髓损伤用户的广泛研究来解决CPS的工程问题。

CPS: Synergy: Collaborative Research: Learning control sharing strategies for assistive cyber-physical systemsAssistive machines - like powered wheelchairs, myoelectric prostheses and robotic arms - promote independence and ability in those with severe motor impairments. As the state- of-the-art in these assistive Cyber-Physical Systems (CPSs) advances, more dexterous and capable machines hold the promise to revolutionize ways in which those with motor impairments can interact within society and with their loved ones, and to care for themselves with independence. However, as these machines become more capable, they often also become more complex. Which raises the question: how to control this added complexity? A new paradigm is proposed for controlling complex assistive Cyber-Physical Systems (CPSs), like robotic arms mounted on wheelchairs, via simple low-dimensional control interfaces that are accessible to persons with severe motor impairments, like 2-D joysticks or 1-D Sip-N-Puff interfaces. Traditional interfaces cover only a portion of the control space, and during teleoperation it is necessary to switch between different control modes to access the full control space. Robotics automation may be leveraged to anticipate when to switch between different control modes. This approach is a departure from the majority of control sharing approaches within assistive domains, which either partition the control space and allocate different portions to the robot and human, or augment the human's control signals to bridge the dimensionality gap. How to best share control within assistive domains remains an open question, and an appealing characteristic of this approach is that the user is kept maximally in control since their signals are not altered or augmented. The public health impact is significant, by increasing the independence of those with severe motor impairments and/or paralysis. Multiple efforts will facilitate large-scale deployment of our results, including a collaboration with Kinova, a manufacturer of assistive robotic arms, and a partnership with Rehabilitation Institute of Chicago. The proposal introduces a formalism for assistive mode-switching that is grounded in hybrid dynamical systems theory, and aims to ease the burden of teleoperating high-dimensional assistive robots. By modeling this CPS as a hybrid dynamical system, assistance can be modeled as optimization over a desired cost function. The system's uncertainty over the user's goals can be modeled via a Partially Observable Markov Decision Processes. This model provides the natural scaffolding for learning user preferences. Through user studies, this project aims to address the following research questions: (Q1) Expense: How expensive is mode-switching? (Q2) Customization Need: Do we need to learn mode-switching from specific users? (Q3) Learning Assistance: How can we learn mode-switching paradigms from a user? (Q4) Goal Uncertainty: How should the assistance act under goal uncertainty? How will users respond? The proposal leverages the teams shared expertise in manipulation, algorithm development, and deploying real-world robotic systems. The proposal also leverages the teams complementary strengths on deploying advanced manipulation platforms, robotic motion planning and manipulation, and human-robot comanipulation, and on robot learning from human demonstration, control policy adaptation, and human rehabilitation. The proposed work targets the easier operation of robotic arms by severely paralyzed users. The need to control many degrees of freedom (DoF) gives rise to mode-switching during teleoperation. The switching itself can be cumbersome even with 2- and 3-axis joysticks, and becomes prohibitively so with more limited (1-D) interfaces. Easing the operation of switching not only lowers this burden on those already able to operate robotic arms, but may open use to populations to whom assistive robotic arms are currently inaccessible. This work is clearly synergistic: at the intersection of robotic manipulation, human rehabilitation, control theory, machine learning, human-robot interaction and clinical studies. The project addresses the science of CPS by developing new models of the interaction dynamics between the system and the user, the technology of CPS by developing new interfaces and interaction modalities with strong theoretical foundations, and the engineering of CPS by deploying our algorithms on real robot hardware and extensive studies with able-bodied and users with sprinal cord injuries.