CPS: Medium: Computation-Aware Autonomy for Timely and Resilient Multi-Agent Systems

我们正在进入一个前所未有的信息获取时代，在这个时代，变革的方法正在展示一个清晰的愿景，即自主驱动的未来。自动驾驶汽车、精准农业、机器人监控和基础设施检查只是经历自动驾驶革命的几个领域。为了在这个充满希望的方向上继续前进，我们必须促进各种网络物理系统（CPS）的安全可靠协调。不幸的是，目前我们对这一目标的理解存在很大的差距：决策、传感和运动算法与底层计算资源之间存在明显的鸿沟。因此，本项目试图通过回答以下问题来定义计算感知自主性：(1)环境如何影响计算？(2)自治应该如何适应提高计算意识？(3)如何在运行时优化计算资源以支持自治？(4)自治软件是如何适应错误的？该项目旨在通过优化、计算资源管理和软件弹性来回答这些问题，并在室外机器人试验台进行评估。最后，这项工作的更广泛的影响包括：(1)与弗吉尼亚理工大学工程多样性增强中心合作，为代表性不足的学生提供K-12学术经验；(2)自主课程与设计项目；(3)通过国际无人驾驶车辆系统协会的山脊和山谷分会参加一系列研讨会。本项目主要研究：(1)基于运行时资源优化和软件可靠性约束的多智能体任务分配和运动规划的统一理论和可扩展算法；(2)时间关键型CPS运行时资源优化的高效分析与优化技术；(3)在自治计算核中实现软错误弹性的轻量级和灵活方法；(4)针对目标跟踪和基础设施映射任务的异构多智能体测试平台。该项目将在计算和可靠性感知任务分配、运动规划、运行时资源优化和灵活软件可靠性等大量未开发领域推进知识。新的统一方法完成了鲁棒任务分配和运动规划的循环，因为它作为一种基本工具，在动态环境中日益复杂的多代理任务中，提高CPS的可扩展性、适应性、弹性、安全性、安全性和可用性，并具有可证明的行为。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。

We are entering an age of unprecedented access to information, where transformational methodologies are demonstrating a clear vision of an autonomy-driven future. Self-driving cars, precision agriculture, robotic monitoring, and infrastructure inspection are but a few areas experiencing an autonomy revolution. To continue in this promising direction, it is critical that we facilitate the safe and reliable coordination of diverse cyber-physical systems (CPS).Unfortunately, at present there is a wide gap in our understanding that limits this goal: a stark divide exists between algorithms for decision-making, sensing, and motion, and underlying computational resources. This project therefore seeks to define computation-aware autonomy by answering the following questions: (1) How does an environment impact computation? (2) How should autonomy adapt to improve computational awareness? (3) How are computational resources optimized at run-time in support of autonomy? and (4) How is autonomy software rendered resilient to errors? This project aims to answer these questions through optimization, computational resource management, and software resilience, with evaluation in an outdoor robotic testbed. Finally, the broader impacts of this work include: (1) K-12 academic experiences for underrepresented students in collaboration with Virginia Tech's Center for Enhancement of Engineering Diversity; (2) autonomy curriculum and design projects; and (3) participation in a series of symposiums through the Ridge and Valley chapter of the Association for Unmanned Vehicle Systems International.This project focuses on the investigation of: (1) a unified theory and scalable algorithms for multi-agent task allocation and motion planning with constraints on run-time resource optimization and software reliability; (2) efficient analysis and optimization techniques for run-time resource optimization in time-critical CPS; (3) lightweight and flexible methodologies for achieving soft error resilience in computational kernels for autonomy; and (4) a heterogeneous multi-agent testbed for target tracking and infrastructure mapping missions. This project will advance knowledge in the largely unexplored areas of computation and reliability-aware task allocation and motion planning, run-time resource optimization, and flexible software reliability. The new unified approach closes the loop for robust task allocation and motion planning as it acts as a fundamental tool to advance scalability, adaptability, resiliency, safety, security, and usability of CPS with provable behaviors in increasingly complex multi-agent missions across dynamic environments.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.