CRII: RI: Responsive Movement Primitives for Soft Robot Control

控制基于许多传感器的许多执行器对于开发模仿人类能力的智能机器人、能够去人类无法去的地方的移动机器人以及一般的软机器人（可能是可穿戴的、可吞咽的或在其他微妙环境中安全的）非常重要。动物和人类可以很容易和灵活地对触摸他们皮肤上的任何地方做出反应。因此，例如，一个人可以在黑暗中捡起一把物体，或者一条蠕虫可以在地下洞穴中摸索出路。然而，用于控制机器人传感器和执行器的方法可能无法扩展到更多自由度和传感器。软体机器人尤其不精确，通常是为接触设计的，这意味着需要更强大、更智能的控制。将控制问题分成更简单的子部分，称为原语，可以使控制器的设计和适应更容易。这些原始细胞似乎是生物神经系统的基础。这个项目的目标是为响应的原语建立一种格式，特别是对本地接触感测。这些原语将应用于两种常见类型的机器人问题：操作和运动。该项目是向使机器人具有更逼真的性能和更好地理解生物智能的复杂性迈出的一步。该项目将建立在广泛使用的动态运动原语（DMP）框架上，用连接鞍平衡取代稳定平衡和极限环。鞍平衡将使系统状态的时间随感官输入而变化，从而产生响应运动原语（Responsive Movement Primitives， RMP）。我们将探索RMP的两种设计方法：直接从DMP转换和扩展先前基于鞍座的控制器以获得额外的自由度。第一种方法将用于解决超冗余抓取问题，因为这与发布的DMP基准测试相关。这项研究有可能帮助机器人一次抓取多个物体。第二种方法将用于蠕动蠕虫运动问题，并利用研究者？S前期研究。这项研究将帮助工程师更好地理解如何利用软体运动来进入受限空间，并在未来的医学、基础设施访问和搜索和救援中有潜在的应用。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Controlling many actuators based on many sensors is important for developing smart robots that mimic human abilities, mobile robots that go places humans cannot go, and for soft robots in general, which may be wearable, swallowable, or safe for other delicate environments. Animals and humans can easily and deftly respond to touch anywhere on their skin. As a result, for example, a human can pick up a handful of objects in the dark or a worm can feel its way through an underground burrow. However, The approaches that work to control sensors and actuators on robots may not scale up for more degrees of freedom and sensors. Soft robots are especially imprecise and often designed for contact, which means more robust, intelligent control is needed. Separating the control problem into simpler subparts called primitives can make controller design and adaptation easier. Such primitives seem to be fundamental to biological nervous systems. The goal of this project is to establish a format for primitives that are responsive, specifically to local contact sensing. These primitives will be applied to two common types of robot problems: manipulation and locomotion. The project is a step toward enabling more life-like performance in robots and better understanding of the complexity of biological intelligence. This project will build on the widely used Dynamic Movement Primitives (DMP) framework by replacing the stable equilibria and limit cycles with connected saddle equilibria. The saddle equilibria will enable the timing of the system states to vary with sensory input, resulting in Responsive Movement Primitives (RMP). We will explore two design approaches to RMP: converting directly from DMP and extending a previous saddle-based controller to additional degrees of freedom. The first approach will be implemented for a hyper-redundant grasping problem because this relates to published DMP benchmarks. This study has the potential to help robots pick up multiple objects in a single grasp. The second approach will be implemented for peristaltic worm-like locomotion problem and leverages the investigator?s prior research. This study will help engineers better understand how soft body locomotion can be used to access constrained spaces with potential future application in medicine, infrastructure access, and search and rescue.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.