NRI: Collaborative Research: Efficient Algorithms for Contact-Aware State Estimation

NRI：协作研究：接触感知状态估计的有效算法

• 批准号: 1426703
• 负责人: Michael Kaess
• 金额: $ 40万
• 依托单位: Carnegie-Mellon University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2014
• 资助国家: 美国
• 起止时间: 2014-08-15 至 2018-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1426703&HistoricalAwards=false
• 关键词: NRI; Collaborative; Research; Efficient; Algorithms

This project addresses the difficult theoretical, computational, and applied challenges required to exploit a deep mathematical relationship between recent advances in machine perception/estimation algorithms and recent advances in algorithms for planning/controlling systems undergoing frictional contact. It explores the immediate applications of these algorithms to perception, robotic object manipulation and parts assembly, and humanoid robots performing complex, multi-contact, whole-body maneuvers. In order to showcase the generality of approach, and simultaneously reach out to the important under-represented minority population, the research team employs a new hands-on short-course curriculum in which students apply the proposed algorithms to predict the outcome of games using visual tracking. The course is being developed in partnership with the MIT Office of Minority Education (OME).The project brings together expertise in simultaneous localization and mapping (SLAM), robot manipulation, robotic automation, legged robots, and optimization and nonlinear control, leading to a cross-fertilization of ideas and techniques. The research team exploits sparsity in the complementarity formulations of contact in Lagrangian dynamics. The project explores a new algebraic approach to nonlinear estimator design. The project produces new theorems, new algorithms, and experimental results on real robots. The project also represents a new partnership with our industrial collaborator, ABB Robotics. The developed algorithms facilitate a broad range of new applications in which perception and control systems monitor and manipulate physical interactions with the world. From palm-sized smart devices to environmental monitoring, sensors are becoming ubiquitous; to reach their full potential these sensor networks must be able to reason about contact - the basic building block of physical interaction.

该项目解决了利用机器感知/估计算法的最新进展与经历摩擦接触的规划/控制系统的算法的最新进展之间的深层数学关系所需的困难的理论、计算和应用挑战。 它探索了这些算法在感知、机器人对象操纵和零件组装以及执行复杂、多接触、全身操作的人形机器人方面的直接应用。为了展示方法的通用性，同时接触到重要的未被充分代表的少数群体，研究团队采用了一种新的实践短期课程，学生可以应用所提出的算法来使用视觉来预测游戏结果追踪。该课程是与麻省理工学院少数族裔教育办公室 (OME) 合作开发的。该项目汇集了同步定位和建图 (SLAM)、机器人操纵、机器人自动化、腿式机器人以及优化和非线性控制方面的专业知识，从而形成了思想和技术的交叉融合。 研究小组利用拉格朗日动力学中接触互补公式的稀疏性。 该项目探索了一种新的非线性估计器设计代数方法。该项目产生了新定理、新算法以及在真实机器人上的实验结果。 该项目还代表了我们与工业合作伙伴 ABB Robotics 的新合作伙伴关系。开发的算法促进了广泛的新应用，其中感知和控制系统监视和操纵与世界的物理交互。 从手掌大小的智能设备到环境监测，传感器正变得无处不在。为了充分发挥其潜力，这些传感器网络必须能够推理接触——物理交互的基本组成部分。