基于MATD3算法的多智能体避碰控制

使用多智能体双延迟深度确定性策略梯度(Multi-agent Twin Delayed Deep Deterministic Policy Gradient,MATD3)算法研究了多无人机的避障和到达目标点问题,首先,利用MATD3算法的优越性提高训练效率。其次,基于人工势场法的思想设计了稠密碰撞奖励函数,使得智能体在没有找到最优解决方案时也能得到积极的反馈,加快学习速度。最后,在仿真实验阶段,通过设计的三组对比实验和泛化实验验证了算法的有效性。     

具有群集行为的时变函数分布式优化

本文对具有群集行为的连续时间多智能体系统的优化问题进行了研究. 考虑具有二阶动力学的多智能体系统, 每个智能体都具有一个局部的时变代价函数. 本文的目标是仅仅依靠局部信息交流使得多智能体在运动的过程中保持 连通性、避免碰撞、总体代价函数最小. 为此本文设计了一种具有群集行为的控制协议, 该协议仅仅依赖于自己和邻居 的速度. 可以证明在该控制协议作用下, 所有智能体在保持连通、避免碰撞的同时, 速度能够跟踪上最优速度. 最后, 通 过一个仿真来说明本文的结果.     

融合沿墙行为的多移动机器人有序化多模态群集运动控制

一类基于势场原理的群集控制理论正逐步应用于多agent(智能体)/机器人稳定协同运动中.针对群集运动系统在非规则障碍物环境中运行时易出现的局部极小问题,引入基于行为的机器人学理念,构成多移动机器人多模态群集控制系统.在此框架内,仿生的动物沿端行为与有序化群集运动控制策略相融合,实现了多移动机器人系统快速聚合行为与高效避障行为的统一.移动机器人仿真实验验证了该方法的有效性.     

基于模糊人工势场法的多智能体编队控制及避障方法

针对动态环境中多智能体编队控制及避障问题,提出了一种基于模糊人工势场法的编队方法。首先,在领航跟随法的框架下控制编队队形,在动态队形变换策略的异构模式下,使用人工势场法为多智能体编队中每个智能体规划避障路径;其次,利用模糊控制器控制跟随智能体追踪领航智能体,同时保持跟随智能体之间与领航智能体的相对距离,遇到未知障碍物时,及时保持多智能体编队之间的队形并避免碰撞障碍物。针对人工势场法在引力增量系数和斥力增量系数设置的局限性,利用模糊控制器选择出适应环境的增量系数。Matlab仿真实验结果表明,该方法能够有效地解决复杂环境下多智能体编队控制及避障问题,使用效率函数对实验数据进行分析,验证了所优化方法的合理性和有效性。     

多无人机三维群集算法的适用性研究与改进

针对多无人机群三维空间运动的复杂群集控制问题,提出了基于生物群集行为、依据Reynolds规则描述的三维群集控制算法。已有的研究大多将无人机群集运动简化为二维平面运动,但这不符合实际控制需求。为此,将群集控制算法和人工势场算法推广到三维无人机群集控制中,建立了三维无人机群空间运动模型,通过多种不同条件下的仿真,研究了两种算法在三维群集控制中的有效性。结果显示两种算法用于三维群集控制均具有一定效果,但相对二维所需要的条件更为苛刻。同时,注意到智能算法具有更好的群体聚集效果,而人工势场算法则避碰效果更迅速明显。据此,对人工势场算法和智能算法进行了改进,通过在距离大于平衡点时采用智能算法聚集,在距离小于平衡点时采用人工势场算法避碰,得到能同时获得更好的聚集、避碰效果的新的群集控制算法。     

多智能体同时到达多目标点的协同强化学习算法

针对动态未知环境下多智能体多目标协同问题,为实现在动态未知环境下多个智能体能够同时到达所有目标点,设计函数式奖励函数,对强化学习算法进行改进.智能体与环境交互,不断重复"探索-学习-决策"过程,在与环境的交互中积累经验并优化策略,在未预先分配目标点的情况下,智能体通过协同决策,能够避开环境中的静态障碍物和动态障碍物,同时到达所有目标点.仿真结果表明,该算法相比现有多智能体协同方法的学习速度平均提高约42.86％,同时智能体能够获得更多的奖励,可以做到自主决策自主分配目标,并且实现同时到达所有目标点的目标.     

融合局部自适应追踪的多目标牵制蜂拥算法

针对以往的多智能体蜂拥控制算法在考虑单个目标追踪情形时不具普适性,以及现有的多目标蜂拥控制都是基于全局目标信息来进行集中式协调控制,而非基于局部目标信息下的分布式协调控制的问题,提出一种融合局部自适应检测机制的分布式协同牵制蜂拥算法。首先,算法在分离、聚合、速度匹配和引导反馈的基础上,引入局部自适应追踪策略,实现智能体的局部动态跟随运动;其次,受牵制思想启发,根据节点影响力指数评估算法选取m个信息个体分别向m个目标进行多目标追踪,起到模拟外部信息的作用,不同的信息个体会由于局部自适应检测机制间接地引领周围局部个体向不同目标进行追踪;最后,设计一类新的聚集和排斥势能函数,实现相同目标智能体的聚集,以及不同目标智能体的避碰,具有可调参数少和效率高的优势。通过三维仿真实验验证了算法的多目标追踪可行性和有效性。     

面向未知环境及动态障碍的人工势场路径规划算法

实际战场环境错综复杂,很多隐蔽、动态的障碍无法通过高空手段预先探测得知,因而对智能体执行任务的安全性产生威胁。针对未知且障碍形态多样的战场环境,以躲避动、静障碍,追踪目标为研究对象,提出一种面向未知环境及动态障碍的改进人工势场(Artificial Potential Field,APF)路径规划算法。在该算法中,智能体构建了以目标点为中心的引力势场,以及以障碍物为中心的斥力势场,在智能体行进路途中感知局部障碍及目标点的运动信息,并且将信息加入势场函数的计算中达到动态避障与追踪的效果;另一方面,引入距离因子及动态临时目标点来消除APF算法常见的无解问题——极小解情况及路径抖动现象。通过建立不同数量的随机障碍场景,进行多次仿真对比实验,结果表明:所提算法能够在未知环境中灵活躲避动态障碍并进行目标点的追踪,可以有效消除死解及路径抖动问题。将所提算法与传统APF算法及添加了动态避障机制的文献[19]所述算法进行对比实验,结果表明所提算法能成功化解两种对比算法路径规划失败的情况,顺利完成路径规划任务,且成功率在95%以上。     

多智能体系统的自适应群集分布式优化

本文对具有非线性函数群集行为的连续时间多智能体系统的分布式优化问题进行了研究。本文的目的是 使局部代价函数之和最小。每个智能体只知道与其对应的代价函数。为了解决这一问题,本文设计了一个分布式 控制律,在这个研究中该控制律仅仅依赖于自己和邻居的速度。通过李雅普诺夫稳定性证明了多智能体系统的收 敛性,而且在最小化局部代价函数之和的同时所有智能体可以避免碰撞。最后,通过一个仿真案例来说明所获得 的分析结果。     

具有群集行为的时变函数分布式优化（英文）

本文对具有群集行为的连续时间多智能体系统的优化问题进行了研究.考虑具有二阶动力学的多智能体系统,每个智能体都具有一个局部的时变代价函数.本文的目标是仅仅依靠局部信息交流使得多智能体在运动的过程中保持连通性、避免碰撞、总体代价函数最小.为此本文设计了一种具有群集行为的控制协议,该协议仅仅依赖于自己和邻居的速度.可以证明在该控制协议作用下,所有智能体在保持连通、避免碰撞的同时,速度能够跟踪上最优速度.最后,通过一个仿真来说明本文的结果.     

A local flocking algorithm of multi-agent dynamic systems

In this paper, the local flocking of multi-agent systems is investigated, which means all agents form some groups of surrounding multiple targets with the partial information exchange. For the purpose of realising local multi-flocking, a control algorithm of local flocking is proposed, which is a biologically inspired approach that assimilates key characteristics of flocking and anti-flocking. In the process of surrounding mobile targets through the control algorithm, all agents can adaptively choose between two work modes to depend on the variation of visual field and the number of pursuing agents with the mobile target. One is a flocking pursuing mode which is that some agents pursue each mobile target, the other is an anti-flocking searching mode that means with the exception of the pursing agents of mobile targets, other agents respectively hunt for optimal the mobile target with a closest principle between the agent and the target. In two work modes, the agents are controlled severally via the different control protocol. By the Lyapunov theorem, the stability of the second-order multi-agent system is proven in detail. Finally, simulation results verify the effectiveness of the proposed algorithm.     

邻域交互结构优化的多智能体快速蜂拥控制算法

针对多智能体系统在动态演化过程中容易出现的"局部聚集"现象,融 合复杂网络中的拓扑结构优化理论与多智能体系统协调蜂拥控制研究,提出了一种基 于邻域交互结构优化的多智能体快速蜂拥控制算法.该算法首先从宏观上分析多智 能体的局部聚集现象,利用社团划分算法将局部相对密集的多个智能体聚类成一个 社团,整个多智能体系统可以划分成多个相对稀疏的社团,并为每个社团选择度 最大的个体作为信息智能体,该个体可以获知虚拟领导者信息;随后从多智能体 系统中不同社团相邻个体间的局部交互结构入手,取消社团间相邻个体的交 互作用,设计仅依赖于社团内部邻居个体交互作用的蜂拥控制律;理论分 析表明,只要每个社团存在一个信息智能体,在虚拟领导者的引导作用下,整个多 智能体系统就可以实现收敛的蜂拥控制行为;仿真实验也证实了对多智 能体系统进行邻域交互结构优化可以有效提高整个系统的收敛速度.     

Target tracking and obstacle avoidance for multi-agent systems

This paper considers the problems of target tracking and obstacle avoidance for multi-agent systems. To solve the problem that multiple agents cannot effectively track the target while avoiding obstacle in dynamic environment, a novel control algorithm based on potential function and behavior rules is proposed. Meanwhile, the interactions among agents are also considered. According to the state whether an agent is within the area of its neighbors' influence, two kinds of potential functions are presented. Meanwhile, the distributed control input of each agent is determined by relative velocities as well as relative positions among agents, target and obstacle. The maximum linear speed of the agents is also discussed. Finally, simulation studies are given to demonstrate the performance of the proposed algorithm.     

Flocking for multi-agent systems with optimally rigid topology based on information weighted Kalman consensus filter

This paper investigates the leader-follower flocking problem of multi-agent systems. The leader with input noise is estimated by a proposed continuous-time information weighted Kalman consensus filter (IWKCF) for agents. A novel distributed flocking algorithm based on the IWKCF is further presented to make agents achieve flocking to the leader. It is shown that the proposed flocking algorithm based on the continuous-time IWKCF is asymptotically stable. Applying the topology optimization scheme, the communication complexity of system topologies of multi-agent systems is effectively reduced. Finally, simulations are provided to demonstrate the effectiveness of the proposed results.     

Flocking of Multi-Agents With a Virtual Leader

All agents being informed and the virtual leader traveling at a constant velocity are the two critical assumptions seen in the recent literature on flocking in multi-agent systems. Under these assumptions, Olfati-Saber in a recent IEEE Transactions on Automatic Control paper proposed a flocking algorithm which by incorporating a navigational feedback enables a group of agents to track a virtual leader. This paper revisits the problem of multi-agent flocking in the absence of the above two assumptions. We first show that, even when only a fraction of agents are informed, the Olfati-Saber flocking algorithm still enables all the informed agents to move with the desired constant velocity, and an uninformed agent to also move with the same desired velocity if it can be influenced by the informed agents from time to time during the evolution. Numerical simulation demonstrates that a very small group of the informed agents can cause most of the agents to move with the desired velocity and the larger the informed group is the bigger portion of agents will move with the desired velocity. In the situation where the virtual leader travels with a varying velocity, we propose modification to the Olfati-Saber algorithm and show that the resulting algorithm enables the asymptotic tracking of the virtual leader. That is, the position and velocity of the center of mass of all agents will converge exponentially to those of the virtual leader. The convergent rate is also given.      

Target tracking and obstacle avoidance for multi-agent networks with input constraints

In this paper, the problems of target tracking and obstacle avoidance for multi-agent networks with input constraints are investigated. When there is a moving obstacle, the control objectives are to make the agents track a moving target and to avoid collisions among agents. First, without considering the input constraints, a novel distributed controller can be obtained based on the potential function. Second, at each sampling time, the control algorithm is optimized. Furthermore, to solve the problem that agents cannot effectively avoid the obstacles in dynamic environment where the obstacles are moving, a new velocity repulsive potential is designed. One advantage of the designed control algorithm is that each agent only requires local knowledge of its neighboring agents. Finally, simulation results are provided to verify the effectiveness of the proposed approach.     

Second-order tracking control for leader-follower multi-agent flocking in directed graphs with switching topology

This paper investigates the flocking problem for leader-follower multi-agent systems in directed graphs with switching topology. A decentralized state control rule, namely, a second-order protocol, is designed for each agent to track the leader. And it is proved that the proposed control scheme can effectively estimate the tracking error of each agent when the leader is active. Particularly, to ensure the tracking error can be estimated, the following two questions are solved: (1) How many agents are needed to connect to the leader? (2) How should these connections be distributed? Finally, a simple example is also given to verify the effectiveness of the proposed theorems.     

Flocking with connectivity preservation for disturbed nonlinear multi-agent systems by output feedback

This paper considers the problem of flocking with connectivity preservation for a class of disturbed nonlinear multi-agent systems. In order to deal with the nonlinearities in the dynamic of all agents, some auxiliary variables are introduced into the state observer for stability analysis. By proposing a bounded potential function and using adaptive theory, a novel output feedback consensus algorithm is developed to guarantee that the states of all agents achieve flocking with connectivity preservation.     

Dynamic target tracking and observing in a mobile sensor network

This paper presents novel approaches to (1) the problem of flocking control of a mobile sensor network to track and observe a moving target and (2) the problem of sensor splitting/merging to track and observe multiple targets in a dynamic fashion. First, to deal with complex environments when the mobile sensor network has to pass through a narrow space among obstacles, we propose an adaptive flocking control algorithm in which each sensor can cooperatively learn the network’s parameters to decide the network size in a decentralized fashion so that the connectivity, tracking performance and formation can be improved. Second, for multiple dynamic target tracking, a seed growing graph partition (SGGP) algorithm is proposed to solve the splitting/merging problem. To validate the adaptive flocking control we tested it and compared it with the regular flocking control algorithm. For multiple dynamic target tracking, to demonstrate the benefit of the SGGP algorithm in terms of total energy and time consumption when sensors split, we compared it with the random selection (RS) algorithm. Several experimental tests validate our theoretical results.