Towards Computationally Efficient One-Shot Design for Performance-Critical Distributed Multi-Agent Control

多智能体系统是由一组分布式交互实体组成的一类重要的互联/网络化系统，它已经成为航天器编队飞行、空中交通管理、传感器网络等各种工程应用的一个强有力的范例。通过协作，多智能体系统可以完成许多超越单个动力系统能力的复杂控制任务，例如移动超大物体、环境监测和灾难搜索/救援。此外，多智能体系统可以更快地解决一些问题，使用并行性和增加冗余的鲁棒性。然而，实现合作的多智能体系统也提出了挑战，使许多相关的应用程序（特别是那些要求关键的控制性能）仍然是概念。一个重要的挑战在于缺乏一个系统的方法，允许控制工程师处理相关的多代理分布式控制设计在计算效率和完全集成的方式。现有的方法分离的设计成高，低层次的控制往往无法保证可靠性，这是一个关键的要求，接受控制工程师。该项目支持基础研究，以提供克服这些挑战所需的知识，从而促进多代理分布式控制技术在现实世界中的更广泛应用。该项目还将通过开发跨部门的机器人工程计划来促进工程教育，并促进少数民族参与科学研究。该项目的目标是通过以下方式为分布式多智能体控制理论的发展做出根本性贡献：（i）开发新颖的混合切换控制方案，以解决复杂因素（例如，代理的物理动力学，致动和数据采样的限制，通信延迟）在一个整体的，一次性的分布式控制设计，以及（ii）生成有效的计算工具，从结合确定性和概率的角度，使平衡设计的复杂性和控制性能。它将为该领域引入创新的方法和工具，导致以下重要的范式变化：（i）从目前研究中主要采用的分离的两步设计到具有可证明的网络稳定性和可控性能的整体一次性设计;（二）从主要用于当前稳定性分析和分布式控制综合的简单二次型的李雅普诺夫函数到具有先进的复合形式，将显着降低分析保守性和提高控制性能;以及（iii）从现有方法主要利用的具有平凡状态/输出反馈控制器结构的分布式算法到具有混合连续时间和离散事件动态的新型混合控制器结构的分布式算法，便于简化的分布式最优控制合成，线凸优化。该奖项反映了NSF的法定使命，并已被认为是值得通过评估使用基金会的智力价值的支持和更广泛的影响审查标准。

Multi-agent systems, an important class of interconnected/networked systems composing a group of distributed interacting entities, have been emerging as a powerful paradigm for various unprecedented engineering applications, such as spacecraft formation flying, air-traffic management, sensory networks, etc. Through collaboration, a multi-agent system can accomplish numerous complicated control tasks that surpass the capability of a single dynamical system, such as moving an oversized object, environmental monitoring, and disaster search/rescue. Moreover, a multi-agent system can solve some problems faster using parallelism and increase robustness through redundancy. However, implementing cooperative multi-agent systems also presents challenges, making many related applications (especially those demanding critical controlled performance) remain conceptual. One important challenge lies in the lack of a systematic approach that allows control engineers to treat associated multi-agent distributed control design in a computationally-efficient and fully-integrated manner. Existing approaches of separating the design into high- and low-level controls often fail to analytically guarantee reliability, which is a critical requirement for acceptance by control engineers. This project supports fundamental research to provide the knowledge needed to overcome these challenges, thereby promoting broader real-world applications of multi-agent distributed control techniques. This project will also create unique opportunities to promote engineering education through the development of a cross-departmental robotics engineering program, and to boost minority involvement in scientific research.The goal of this project is to make fundamental contributions to the advancement of distributed multi-agent control theory by (i) developing novel hybrid switching control schemes to address complicated factors (e.g., agent’s physical dynamics, actuation and data sampling limitations, communication delays) in a holistic, one-shot distributed control design, and (ii) generating effective computational tools from combined deterministic and probabilistic perspectives to enable balancing design complexity and controlled performance. It will introduce innovative methodologies and tools to the field, leading to the following important paradigm changes: (i) from separated two-step designs dominantly adopted in current study to holistic, one-shot designs with provable network stability and controlled performance; (ii) from Lyapunov functions with simple quadratic forms dominantly utilized for current stability analysis and distributed control synthesis to Lyapunov functions with advanced composite forms that would significantly reduce analysis conservatism and improve controlled performance; and (iii) from distributed algorithms with trivial state/output feedback controller structures dominantly exploited by existing methods to distributed algorithms with novel hybrid controller structures of mixed continuous-time and discrete-event dynamics, facilitating simplified distributed optimal control synthesis via off-line convex optimization.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.