Formation Control of Multi-Agent Systems using Continuum Models

本项目旨在开发基于连续体模型的互联多智能体系统分散编队规划和编队控制的新方法。因此，与经典的离散方法相反，其中多智能体系统的动力学是由通信拓扑的规范和交换数据的相对权重决定的，全局系统动力学是根据偏（积分）微分方程先验地施加的。与基于空间和时变密度分布的欧拉视角不同，该设置分别在代理位置或速度中使用显式公式，独立的空间坐标表示连续的通信路径。这种多智能体连续体的分布参数描述可以作为控制设计的基础，以实现沿着规定的过渡路径在有限时间内实现地层剖面之间的过渡，同时考虑到避免碰撞。为此，该项目将侧重于基于平面度的运动规划方法的扩展和适当组合，以及用于稳定跟踪误差动力学的前馈控制和后退方法。考虑连续体模型意味着设计独立于实际通信拓扑。后者是通过适当的结构保持离散化技术，直接实现控制设计和各自的闭环动力学在相互连接的个体代理的水平。开发的方法将分别基于执行器和传感器网络或群体的仿真研究进行评估，并在移动机器人网络中进行实验验证。

This project aims at developing new methods for decentralized formation planning and formation control of interconnected multi-agent systems based on continuum models. With this, contrary to the classical discrete approach, where the dynamics of the multi-agent system is determined by the specification of the communication topology and the relative weighting of the exchange data, the global system dynamics is a priori imposed in terms of partial (integro-) differential equations. Differing from the Eulerian perspective which is based on spatial and time-varying density distributions the proposed setting makes use of an explicit formulation in the agent position or velocity, respectively, with the independent spatial coordinate representing the continuous communication path. This distributed-parameter description of a multi-agent continuum serves as basis for control design with the desire to achieve finite time transitions between formation profiles along prescribed transition paths taking into account collision avoidance. For this, the project will focus on the extension and suitable combination of flatness-based methods for motion planning and feedforward control and backstepping methods for the stabilization of the tracking error dynamics. The consideration of continuum models implies independence of the design from the actual communication topology. The latter is imposed by proper structure-preserving discretization techniques which directly enable to realize the control design and the respective closed-loop dynamics at the level of interconnected individual agents. The developed methods will be evaluated based on simulation studies for actuator and sensor networks or swarms, respectively, as well as validated experimentally for a network of mobile robots.