Collaborative Research: Self-Identification for Robot Manipulation under Uncertainty Aided by Passive Adaptability

这笔赠款支持将在机器人操纵和主动感知的交叉点上贡献新知识的研究，促进科学进步和国家繁荣。要使通用机器人能够与世界进行复杂的物理交互，机器人在有限感知下工作的能力是必不可少的。然而，由于传统的方法将感知和操作顺序地描述为解耦的系统组件，机器人的操作技能已经被动地受到感知系统的约束。该奖项支持建立一种新的范式来实现感知和操纵之间的交互作用的研究，并将从根本上改变两者的角色，以积极促进彼此。关键的概念被称为自我识别，这是一个过程，在这个过程中，机器人在对系统甚至自身没有完全了解的情况下开始操纵物体，同时为感知组件创造机会来获取必要的信息，否则这些信息是不可能的。反过来，操作能力随着获得的额外信息而显著升级。由于这一新能力可以改善许多现实世界中的机器人应用，如工业生产、家庭服务和医疗保健应用，这项研究的结果将使美国经济和社会受益。这项研究涉及多个主题，从计算机科学、机械工程和传感器技术，到控制理论和人工智能。多学科框架将扩大未被充分代表的群体的参与，并对工程教育产生积极影响。该项目利用机器人机械手的各种被动(或低水平)适应性，特别是来自机械顺应性(弹簧或软结构)、低级阻抗控制或欠驱动机构的适应性，使机器人能够进行外部观察的探索性运动，并在自适应估计方案中使用，以自我识别系统。机器人-对象-环境系统将被主动重构，同时以基本开环的方式保持期望的状态和稳定性，而这些变化的外部观测被用于生成在线估计和控制器。从本质上讲，它改变了传统的“感觉、计划、行动”范式，转向“行动、感觉、计划”，强调在有限感知下高效地执行任务。研究小组将为模型完全、模型不完整和非模型操作系统建立通用的自我识别框架，以：1)在需要时补偿有限的感知能力，以完成传统上不可行的任务；2)最大限度地提高其感知能力，以进一步提高系统的任务意识和健壮性；3)将自我识别的结果或整个过程纳入操作计划和控制中，以实现在物理和感知限制下的健壮操作。该项目由跨部门机器人基础研究计划支持，该计划由工程指导委员会(ENG)和计算机和信息科学与工程指导委员会(CEISE)联合管理和资助。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

This grant supports research that will contribute novel knowledge at the intersection of robotic manipulation and active perception, promoting both the progress of science and the advancement of national prosperity. To enable general-purpose robots that can offer sophisticated physical interactions with the world, the capability of robots to work under limited perception is essential. However, as traditional approaches sequentially formulate perception and manipulation into decoupled system components, robot manipulation skills have been passively constrained by the perception system. This award supports research to establish a new paradigm for enabling the interactions between perception and manipulation, and will fundamentally transform the roles of both to actively facilitate each other. The key concept, termed as self-identification, is a process where robots start to manipulate objects without full knowledge of the system, or even of itself, while in the meantime creating opportunities for the perception component to acquire necessary information that were impossible otherwise. In turn, the manipulation capability is significantly upgraded with the extra information obtained. As this new ability can improve many real-world robot applications, such as industrial production, household services, and healthcare applications, the results from this research will benefit the U.S. economy and society. This research involves several topics ranging from computer science, mechanical engineering, and sensor technology, to control theory and artificial intelligence. The multi-disciplinary framework will broaden the participation of underrepresented groups and positively impact the engineering education.This project leverages various types of passive (or low-level) adaptability in robot manipulators, especially that coming from mechanical compliance (springs or soft structures), low-level impedance control, or underactuated mechanisms, to allow the robot to conduct exploratory motions that are externally observed and used within an adaptive estimation scheme to self-identify the system. The robot-object-environment system will be actively reconfigured while maintaining the desired states and stability in an essentially open-loop way, while external observations of these changes are used to generate online estimations and controllers. Essentially, it changes the traditional paradigm from “sense, plan, act” to “act, sense, plan”, with an emphasis on efficient and effective task execution under limited sensing. The research team will establish generic self-identification frameworks for model-complete, model-incomplete, and model-free manipulation systems to: 1) compensate for limited perception abilities, where needed, to accomplish tasks that are traditionally infeasible; 2) maximize their perception capabilities to further improve the system’s task-awareness and robustness; and 3) incorporate the results or the entire process of self-identification into manipulation planning and control to enable robust manipulation under physical and sensing limitations.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.