RI: Small: Collaborative Research: Extracting Dynamics from Limited Data for Modeling and Control of Unmanned Autonomous Systems

自主机器人可以为许多现实世界的应用做出贡献，例如紧急响应和搜索与救援。然而，对于某些应用来说，部署大型机器人的成本可能太高，而小型机器人是首选。这种小型机器人的一个共同问题是在不确定性存在的情况下可靠地导航。目前的机器人建模和控制方法无法捕捉小尺度不确定性影响所带来的复杂性。此外，小尺寸限制了传感器和计算有效载荷，这限制了机器人的感知和控制能力。该项目引入了一个数据驱动的建模框架，通过控制小型机器人的可靠导航来量化和利用不确定性。该项目使本科生能够参与研究，并利用加州大学河滨分校（UC Riverside）这个为西班牙裔服务的机构的学生多样性，扩大代表性不足群体的参与。本项目研究机器人与环境相互作用中的不确定性影响机器人行为的机制。小型机器人的运动更具随机性，因为执行器的误差和与环境的不确定相互作用放大了姿态误差。目标是引入一个平台无关的、数据驱动的建模框架来量化不确定性，并随后通过控制来利用它，在不确定性下进行可靠的机器人导航。具体目标是：1)利用有限的数据提取动力学，为不确定系统建模；2)基于派生的降阶模型综合基于不确定性感知模型的控制器；3)用空中、地面和海洋机器人测试和验证理论分析、推导模型和控制算法。光谱方法用于提取时空动态和量化不确定性。一种模型参考自适应控制方案利用提取的动力学和不确定性来实现机器人的可靠导航。虽然本研究中开发的基本原理是基于小型机器人的，但这个项目的发现可能会推广到具有有限传感和噪声驱动的大型机器人。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。

Autonomous robots can contribute to many real-world applications, such as emergency response and search-and-rescue. For some applications, however, the cost of deploying large robots may be too high, and small robots are preferred. A common issue with such small robots is navigating reliably in the presence of uncertainty. Current robot modeling and control approaches cannot capture the intricacies imposed by the effect of uncertainty at small scales. In addition, the small size restricts sensor and computational payloads, which limit the robot's perceptual and control capabilities. This project introduces a data-driven modeling framework to quantify and exploit uncertainty via control for reliable navigation of small robots. The project enables undergraduate students to become involved in research, and capitalizes on the student diversity at UC Riverside, a Hispanic-serving Institution, to broaden participation of under-represented groups. This project investigates the mechanisms that uncertainties in robot-environment interactions affect robot behavior. Small robot motion is more stochastic since errors at the actuators and uncertain interactions with the environment amplify errors in pose. The goal is to introduce a platform-agnostic, data-driven modeling framework to quantify uncertainty and subsequently exploit it via control for reliable robot navigation under uncertainty. The specific aims are to: 1) extract dynamics using limited data for modeling uncertain systems; 2) synthesize uncertainty-aware model-based controllers based on derived reduced-order models; and 3) test and validate theoretical analysis and derived models and control algorithms with aerial, ground, and marine robots. Spectral methods are used to extract spatio-temporal dynamics and to quantify uncertainty. A model-reference adaptive control scheme utilizes extracted dynamics and uncertainty for reliable robot navigation. While the basic principles developed in this research are grounded on small robots, this project's findings may generalize to larger robots with limited sensing and noisy actuation.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.