Collaborative Research: CPS: Medium: Constraint Aware Planning and Control for Cyber-Physical Systems

协作研究：CPS：中：网络物理系统的约束感知规划和控制

• 批准号: 2039054
• 负责人: Ricardo Sanfelice
• 金额: $ 60万
• 依托单位: University of California-Santa Cruz
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2020
• 资助国家: 美国
• 起止时间: 2020-10-01 至 2024-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2039054&HistoricalAwards=false
• 关键词: Collaborative; Research; CPS; Medium; Constraint

The objective of this work is to generate new fundamental science for computer controlled complex physical systems, a broad class of cyber-physical systems (CPS) and demonstrate this science in aerial vehicles and walking robots. The new science enables autonomous planning and control in the presence of failures and abrupt changes in system variables. A framework for the design of algorithms that exploit awareness of the physical and design constraints to autonomously self-adapt their motion plan and control actions will be generated. The approach exploits elements from geometry, adaptive control, and hybrid control to advance the knowledge on modeling, planning, and design of CPS with constraints, non-smooth, and intertwined continuous and discrete dynamics. Unlike current approaches, which separate the task associated with planning the motion from the design of the algorithm used for control, the algorithms to emerge from this project self-learn and self-adapt in real time to cope with unexpected changes in motion and specification constraints so as to enable autonomous systems to perform robustly and safely, and degrade gracefully under failure conditions. Specifically, the new algorithms will learn and monitor the physical and design constraints in real time and adapt both planner and controller by selecting the appropriate constraints to enforce, with robustness and safety guarantees. The capabilities of the new tools will be demonstrated on multi-legged robots in harsh environments that make them prone to failures, and on aerial vehicles in contested/adversarial environments.The proposed plan contributes to Science of Cyber-Physical Systems by addressing modeling, motion planning, and design of CPS with constraints, non-smooth, and intertwined continuous and discrete dynamics. The merits of the proposal fall into four broad categories: (i) a framework to mathematically formulate learning-based planning and control for CPS with awareness of its constraints, (ii) novel architectures that lead to robust adaptive constraint satisfaction, (iii) deep understanding of roles and priorities of system constraints in CPS, and (iv) tools and design techniques that permit engineers to deploy constraint aware algorithms. The results of this work are broad in their application to all kinds of CPS that require planning and control, in particular, autonomous systems in transportation (air and ground). Synergistic collaborations with researchers at Samsung, the start-up Ghost Robotics, and at the University of Bologna are instrumental in disseminating the application of our results to industry and academia. A synergistic outreach program at UCSC and the University of Michigan impacts high school students and teachers.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.

这项工作的目标是为计算机控制的复杂物理系统（广泛的网络物理系统（CPS））产生新的基础科学，并在飞行器和步行机器人中展示这一科学。这门新科学能够在系统变量出现故障和突然变化的情况下自主规划和控制。将生成一个算法设计框架，该框架利用物理和设计约束的意识来自主地自适应其运动计划和控制动作。该方法利用几何、自适应控制和混合控制的元素来推进具有约束、非光滑、交织连续和离散动力学的CPS的建模、规划和设计知识。与目前的方法不同，这些方法将运动规划与用于控制的算法设计分开，该项目中出现的算法可以实时自我学习和自适应，以应对运动和规范约束的意外变化，从而使自主系统能够稳健安全地执行，并在故障条件下优雅地降级。具体来说，新算法将实时学习和监控物理和设计约束，并通过选择适当的约束来适应规划器和控制器，同时具有鲁棒性和安全性保证。新工具的能力将在恶劣环境中的多腿机器人上进行演示，这些环境使它们容易发生故障，并在竞争/对抗环境中的飞行器上进行演示。提出的计划通过解决具有约束、非光滑和交织的连续和离散动力学的CPS的建模、运动规划和设计，为信息物理系统科学做出了贡献。该提案的优点分为四大类：(i)一个框架，以数学方式制定基于学习的规划和控制，并意识到其约束，（ii）导致鲁棒自适应约束满足的新架构，（iii）对CPS中系统约束的角色和优先级的深刻理解，以及（iv）允许工程师部署约束感知算法的工具和设计技术。这项工作的结果广泛应用于需要规划和控制的各种CPS，特别是运输（空中和地面）的自主系统。与三星（Samsung）、初创企业Ghost Robotics和博洛尼亚大学（University of Bologna）的研究人员的协同合作，有助于将我们的成果应用于工业界和学术界。加州大学圣迭戈分校和密歇根大学的协同推广计划影响了高中学生和教师。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。

项目成果

Uniting Nesterov’s Accelerated Gradient Descent and the Heavy Ball Method for Strongly Convex Functions with Exponential Convergence Rate

结合 Nesterov 的加速梯度下降法和重球法来求解具有指数收敛率的强凸函数

• 发表时间: 2021
• 期刊: Proceedings of the American Control Conference
• 作者: Hustig-Schultz, D.;Sanfelice, R.G.
• 通讯作者: Sanfelice, R.G.

A Self-Triggered Control Strategy to Guarantee Forward Invariance

保证前向不变性的自触发控制策略

• 发表时间: 2021
• 期刊: Proceedings of the American Control Conference
• 作者: Kooi, D.;Sanfelice, R.G.
• 通讯作者: Sanfelice, R.G.

Robust Finite-Time Parameter Estimation for Linear Dynamical Systems

线性动力系统的鲁棒有限时间参数估计

• DOI: 10.1109/cdc45484.2021.9683268
• 发表时间: 2021
• 期刊: Proceedings of the 60th IEEE Conference on Decision and Control
• 作者: Johnson, R.;Saoud, A.;Sanfelice, R.
• 通讯作者: Sanfelice, R.

Hybrid Concurrent Learning for Hybrid Linear Regression

混合线性回归的混合并发学习

• DOI: 10.1109/cdc51059.2022.9992473
• 发表时间: 2022
• 期刊: 2022 IEEE 61st Conference on Decision and Control
• 作者: Johnson, Ryan S.;Sanfelice, Ricardo G.
• 通讯作者: Sanfelice, Ricardo G.

A Totally Asynchronous Block-Based Heavy Ball Algorithm for Convex Optimization

• DOI: 10.23919/acc55779.2023.10156324
• 发表时间: 2023-05
• 期刊: 2023 American Control Conference (ACC)
• 作者: Dawn M. Hustig-Schultz;Katherine R. Hendrickson;M. Hale;R. Sanfelice