CAREER: Exploring Robust Robot Manipulation through Compliance- and Motion-based Manipulation Funnels

该学院早期职业发展（Career）奖支持在非结构化环境中进行通用机器人操作的研究。大多数现实世界的操作任务都涉及不确定性、不可建模的物理和未知参数，传统的精确计划和控制方法已经达到了硬性限制。该奖项支持的研究旨在建立一种新的范式，使机器人能够通过“操纵漏斗”的视角处理不确定性和未知数。操作漏斗的概念与普通使用漏斗相同，其思想是通过由机器人顺应性或运动策略定义的限制性颈部过滤大量任务可能性，以确保随后的机器人动作对不确定性具有鲁棒性。这种新模式将改善现实世界中的机器人应用，例如用于工业生产、家庭服务和医疗保健的机器人应用。该奖项还将支持几项STEM计划，重点是扩大代表性不足群体的参与，包括动手机器人操作教程和配套书籍，提高研究成果的课程，以及为本科生和K-12学生提供研究机会。该项目的目标是通过将任务空间中的几何操作漏斗的思想推广到基于机器人顺应性和运动策略的新型漏斗，从而脱离机器人操作的传统感知、规划和控制管道，以实现对环境不确定性的鲁棒和灵巧操作。在这种情况下，重点是识别条目，塑造颈部，并在这些新的操纵漏斗类中找到出口。例如，通过利用主动或被动顺从，可以主动“打开”最初被阻塞的通道，通过自稳定任务编队精确地操纵对象，并通过扩大规划空间和简化控制来促进丰富接触的操作。同样，通过利用运动和任务约束，可以主动创建通道来及时捕获状态转换，从而有效地减少不确定性，甚至直接找出从不确定操作输入到其可能输出的映射。此外，通过在任务中传递漏斗，并通过漏斗连接组成多模态操作解决方案，所提出的基于漏斗的框架将在确保鲁棒性的同时实现复杂的操作任务。因此，该项目将使机器人能够通过非传统但更可靠的框架进行操作，使它们能够在传统上不可行的高度不确定的情况下工作。该项目由跨部门机器人基础研究项目支持，由工程（ENG）和计算机与信息科学与工程（CISE）联合管理和资助。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。

This Faculty Early Career Development (CAREER) award supports research in general-purpose robotic manipulation in unstructured environments. Most real-world manipulation tasks involve uncertainties, un-modellable physics, and unknown parameters, where traditional approaches for precise planning and control have been hitting a hard limit. This award supports research that seeks to establish a novel paradigm that enables robots to handle uncertainties and unknowns through the lens of “manipulation funnels.” The concept of manipulation funnel is the same as that of an ordinary use funnel, wherein the idea is to filter a large set of task possibilities through a restrictive neck, defined by robot compliance or motion strategy, to a smaller set ensuring that the subsequent robot actions are robust against uncertainties. This new paradigm will improve real-world robot applications, such as those used in industrial production, household services, and healthcare. The award will also support several STEM initiatives, with focus on broadening participation to underrepresented groups, including hands-on robotic manipulation tutorials and an accompanying book, curriculum enhancement with research outcomes, and research opportunities for undergraduate and K-12 students.The objective of this project is to depart from the traditional pipeline of perception, planning, and control for robotic manipulation by generalizing the idea of geometric manipulation funnels in task space to new classes of funnels based on robot compliance and motion strategy for robust and dexterous manipulation against environmental uncertainties. Within this context, the focus is on identifying the entries, shaping the necks, and finding the exits in these new classes of manipulation funnels. For example, by leveraging active or passive compliance, funnels that are initially blocked can be can actively “opened” to precisely manipulate objects through self-stabilizing task formations and facilitate contact-rich manipulation with enlarged planning spaces and simplified control. Similarly, by leveraging motions and task constraints, funnels can be actively created to cage the state transitions in time to effectively reduce uncertainties or even directly figure out the mapping from uncertain manipulation inputs to their possible outputs. Furthermore, by transferring funnels through tasks and composing multi-modal manipulation solutions via funnel concatenations, the proposed funnel-based framework will enable complex manipulation tasks while firmly guaranteeing robustness. As a result, this project will enable robots to manipulate through a non-traditional but more reliable framework, allowing them to work in highly uncertain scenarios that were traditionally infeasible.This project is supported by the cross-directorate Foundational Research in Robotics program, jointly managed and funded by the Directorates for Engineering (ENG) and Computer and Information Science and Engineering (CISE).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.