SIA OpenIR  > 海洋信息技术装备中心
群海洋机器人区域监视方法研究
Alternative TitleStudy on Region Monitoring with a Swarm of Marine Robots
李冠男
Department海洋信息技术装备中心
Thesis Advisor林扬
Keyword海洋机器人 群机器人 区域监视 长期自治 协同任务分配
Pages215页
Degree Discipline机械电子工程
Degree Name博士
2019-11-28
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract我国拥有漫长的海岸线和广阔的海洋国土面积。有效监控海洋国土是宣示主权和国家安全的重要保证。海洋机器人,包括自主水下机器人(Autonomous Underwater Vehicle,AUV)和无人水面艇(Unmanned Surface Vehicle,USV)是区域海域监控的重要工具,随着机器人单体技术的成熟和多机器人协作技术的发展,使用海洋机器人集群对区域水域进行监控成为必然的发展方向。为组织海洋机器人集群对区域水域进行长期持续监视,本文系统地研究了适合于不同集群规模、通信限制和地形是否已知等条件的机器人集群控制策略。主要研究内容如下:(1)基于有向无环图模型的集群区域覆盖监视方法。用于协调控制大规模水下机器人集群,以地毯式覆盖的方式监控水下环境。本方法使用大规模水下机器人集群对区域水域的水下环境进行监视。对区域进行监视时,最理想的情况是传感器的探测范围完全覆盖目标区域。对于水面、陆地和空域,使用卫星、大型雷达等设备可以实现这一目的。而在水下环境中,传感器探测信号衰减快,探测范围有限,难以通过单台设备实现对整个区域的全面监视。为解决这一问题,探索使用大规模海洋机器人集群对水下环境进行覆盖监视。充分发挥集群中机器人个体数量众多的特征,部署机器人形成与目标区域匹配的阵型,实现集群整体探测范围对区域的完全覆盖。因此需要研究集群复杂模式形成问题。这一领域针对水下机器人集群的研究很少,相关研究集中在无人机、陆地移动机器人和无人船集群领域。使用的方法依赖GPS等全局定位设备和使用无线电在集群中构建的低延迟全连接通信网络。通常由特定的主控节点或地面基站对集群中个体进行协调控制,使群体整体形成特定阵型。由于适合于水下机器人集群的低成本通信、探测设备作用距离有限,难以构建全连接的通信网络,机器人只能进行局部通信和相对定位。由于通信受限,地面基站或指定的主控节点难以介入,因此集群需要在没有主控干预的情况下,通过局部通信和相对定位自主部署形成与目标区域匹配的阵型。为解决上述问题,针对两种应用场景对集群部署方法展开研究。针对目标水域地形已知的情况,将集群部署问题映射为一个基于局部感知和通信的大规模集群复杂模式形成问题。基于粒子群优化(PSO)得到机器人在目标水域中的最优分布点云,设计了相应算法将点云映射为树模型或有向无环图模型。基于上述模型,分别研究了基于单先导和双先导两种策略的通信协议与行为规则,以渐进式的方式引导机器人完成部署。使用李雅普诺夫稳定性判据证明了算法的收敛性。针对目标水域地形未知的情况,使用Lennard-Jones虚拟力场驱使集群中的机器人弹性扩散以实现最大程度区域覆盖。机器人在探测过程中要持续消耗能量,以保持自身所处位置,并进行必要的通信、计算等。为避免机器人因为能源耗尽而损失掉,需要使集群在宏观上保持期望阵型以监视区域,同时协调集群中的个体进行能源补给,从而实现集群对区域的持续监视。目前针对这一问题的研究成果非常有限。针对上述问题,给出基于平衡森林拓扑的集群动态能源补给策略并证明其收敛性。为机器人设计了相应的通信协议和行为规则,使集群在保持区域覆盖的情况下形成能源梯度以补给能源,达到能源的动态稳定。通过仿真和移动机器人平台试验对上述研究内容进行了验证。(2)基于BA-CPTD编队的集群区域监视方法。用于控制小规模无人艇集群以编队扫掠的方式对目标水域进行监视。使用无人艇群以编队的方式巡逻监视区域,可以及时发现区域中的异常。由多艘无人艇相互配合对异常情况作出应对,如包围可疑目标、环绕查证目标等,可以提高作业效能,甚至完成单台无人艇难以实现的功能。无人艇群编队控制技术是艇群编队巡逻的基础,快速形成期望队形可以加快艇群对异常情况的反应速度。现有无人艇群编队控制技术通常只保证编队最终收敛,对加快编队收敛速度的研究较少。而时间最优编队控制的方法大都集中在陆地移动机器人领域,通常使用最优控制的思路实现,依赖于系统的精确数学模型。由于无人艇群运动过程中易受水流干扰,难以建立精确的数学模型,因此这类方法难以直接应用于无人艇编队控制中。为解决上述艇群编队形成速度慢的问题,提出一种基于蝙蝠算法(Bat Algorithm)和控制参数化-时间离散化(Control Parameterization & Time Discretization)策略的时间最优编队控制方法,称为BA-CPTD编队控制法。此方法针对海洋机器人运动过程易受海流干扰,难以建立精确数学模型的问题,将系统模型分解为近似数学模型与补偿项的组合。对于近似数学模型部分,通过将系统收敛时间离散化,将控制量参数化为一组向量,从而将编队控制问题建模为一个非线性系统寻优问题。使用蝙蝠算法对此问题进行求解,得到可以将多机器人系统快速引导到期望队形附近的控制律。此后使用基于实时反馈的视线角(Line of Sight)编队控制方法消除由于补偿项引起的编队误差。在此基础上,研究了基于编队扫掠的小规模集群区域监视通信协议与行为规则,实现巡逻过程中队形自主构建,对动态目标的跟踪、包围和查证等。(3)基于动态森林的无人艇集群区域监视方法。控制可以远距离通信的无人艇集群在目标水域中分散巡逻。艇群编队巡逻时无人艇聚集在一起,容易遗漏区域中其他部分的信息。分散巡逻的策略使无人艇均匀散布巡弋整个目标水域,可以及时发现各子区域中的异常,触发艇群对异常情况作出进一步响应,是编队巡逻策略的重要补充。艇群分散巡逻的难点首先在于巡逻区域通常是岛礁、水道等地形不规则的区域,艇群需要均匀分割目标区域以规划出互不重合的闭环巡逻路径。其次无人艇可能因为能源补充、故障损失、任务转换等原因加入或退出巡逻,巡逻策略需要能够自主适应这类集群规模动态变化的情况。最后规划产生的巡逻路径应避免频繁转向,以降低无人艇能耗和控制难度。为解决上述问题,提出在目标水域中构建根据形势动态变化的平衡森林状拓扑,基于此拓扑结构为无人艇规划巡逻路径。按照此思路,分别针对目标水域地形已知和目标水域地形未知两种情况展开研究。在目标水域地形已知的情况下,集群采用集中式拓扑结构,由主控节点根据目标水域的地形将区域均匀分割为若干块子区域,通过构建各子区域的生成树形成森林。为实现目标水域的均匀分割,将问题建模为节点权值图均匀化问题,给出了节点权值图均匀化算法,并基于马尔科夫过程证明算法收敛性。利用节点权值图均匀化算法的变形算法均分目标区域,并构建平衡森林。基于森林状拓扑为各无人艇规划巡逻路径。通过重构平衡森林,使艇群可以自主适应群体规模动态变化。在目标区域地形未知的情况下,基于节点权值图均匀化算法设计合同网协议,动态构建和调整平衡森林,协调艇群探索区域并规划巡逻路径。合同网协议赋予艇群自主适应集群规模动态变化的能力。(4)基于虚拟费洛蒙的集群区域监视方法。利用仿生思路设计行为规则,分别对水下机器人集群分散巡逻策略和无人艇群分散巡逻策略展开研究。现有集群巡逻策略大都使用设计法开发,通过设计复杂的算法和通信协议实现对群体的协调控制。这类思路对机器人的运算能力和通信能力要求较高,限制了方法在海洋环境中的应用,也提高了构建机器人单体的成本,不利于组建大规模集群。自然界中的动物群落仅依赖简单的行为规则和有限的信息交互,就可以完成复杂的行为,表现出群体智能。因此模拟生物群落的行为规则设计机器人集群行为策略,可以使用简单的算法和信息交互策略组织集群。分别针对水下机器人集群和无人艇群展开研究。对于大规模水下机器人集群,针对水下机器人间进行点对点通信实现难度大,成本高的问题,借鉴虫群等群体使用费洛蒙信息进行间接通信这一行为,研究了在水下通信节点辅助下,利用虚拟费洛蒙场协调组织集群进行区域监视的方法。使用基于集群进化的策略动态调整各机器人访问通信节点周期。对于搭载低成本通信设备的无人艇群,针对通信带宽有限这一问题,模仿动物使用费洛蒙标记领地这一行为,研究了基于虚拟费洛蒙领地抢占规则的方法,使集群以较小通信量实现对目标水域的动态分配和巡逻路径规划。
Other AbstractChina has a long coastline and vast marine territory. Monitoring marine territory effectively is important for declaring sovereignty and national security. Marine robots, including autonomous underwater vehicles and unmanned surface vehicles, are important tools for regional monitoring of waters. With the maturity of robotic technology and the development of multi-robot cooperation technology, using marine robot swarms to monitor waters has become an inevitable development direction. In order to organize a marine robot swarm to monitor an area of interest for long-term and continuously, this paper systematically studies swarm control strategies that suitable for different cluster size, communication constraints and whether the terrain is known. The main research contents are as follows: (1)Swarm area coverage monitoring method based on directed acyclic graph model. It is used to coordinate and control large-scale swarms to coverage and monitor the area of interest. In this method, a large-scale underwater robot cluster is used to monitor the underwater environment of regional waters. When monitoring an area, the best situation is that the detection range of the sensor completely covers the target area. For water surface, land and airspace, this can be achieved with the usage of satellite, high power radar and similar equipment. But in the underwater environment, the sensor detection signal attenuation is fast, the detection range is limited, making it difficult to achieve overall monitoring of the whole area through a single device. In order to solve this problem, it is explored to monitor the underwater environment with a large-scale marine robot swarm. As a robot swarm if featured by involving large number of robots, the robots can be deployed to form a formation matching the target area, realizing complete coverage of the area with the detection range of the robot swarm. Therefore, it is necessary to research methods for formation pattern formation with a swarm of robots. In this field, there are few researches focuser on underwater robot robots. Most available methods focus on UAV swarms, UGV swarms and USV swarms. These methods depend on GPS and other global positioning equipment, and a low delay full connection communication network built on radio is usually necessary. A specific master node or ground base station is used to coordinate and control the individuals in the cluster, so that the whole group forms a specific formation. However, in underwater robot swarms, due to the communication constraints of low-cost communication devices and limited working distance of detection equipment, it is difficult to build a fully connected communication network. As a result, robots can only carry out local communication and relative positioning. Thus it is difficult for the ground base station or the designated master node to intervene. Therefore, the robot swarm needs to form a pattern formation matching the target area through local communication and relative positioning automatically, without the intervention of a master node or the ground base station. In order to solve the above problem, the cluster deployment method is studied in two application scenarios. If the terrain of the area of interest is known, the problem is mapped to a swarm pattern formation problem based on local perception and communication. Based on particle swarm optimization (PSO), the point cloud represents the distribution of robots is obtained. Algorithms are proposed to map the point clouds into a tree or a directed acyclic grap. With the above models, communication protocols and behavior rules based on one-predecessor and two-predecessors are studied respectively, leading the robots to join and form the desired pattern in a gradual way. The convergence of the algorithm is proved by Lyapunov stability criterion. Without the knowledge of the terrain, the Lennard-Jones virtual force field is used to drive the robots in the swarm to diffuse according to the terrain to achieve maximum coverage. In the process of detection, the robot needs to consume energy continuously to keep its position and carry out necessary communication and calculation. In order to avoid the loss of robots due to energy depletion, it is necessary to make the cluster maintain the expected formation in the macro to monitor the region, and coordinate the individual in the cluster to get recharged, so as to achieve the continuous monitoring of the region with the robot swarm. At present, there are few research focus on this problem. To solve this problem, the energy management strategy of the swarm is further studied. An energy replenishment strategy based on tree/balanced forest topology is presented and its convergence is proved. Corresponding communication protocols and behavior rules are designed so that the robots will form an energy gradient to replenish energy while maintaining regional coverage, achieving dynamic stability. The above research contents are verified by simulation and experiments with mobile robots. (2) Swarm area monitoring method based on BA-CPTD formation. The method aims at organizing a small-scale USV swarm with the limited communication range to sweeping the area of interest in formation. By using USVs to patrol and monitor the area in formation, the abnormalities in the area can be detected in time. Then multiple USVs can cooperate with each other to respond to abnormal situations, such as surrounding suspicious targets, surrounding verification targets, etc. This can improve the operation efficiency and even carry out tasks that are difficult to be realized by a single unmanned boat. The formation control technology is the basis of the formation patrol of USVs. Accelerating the formation process can speed up the response speed to the abnormal target. Available formation control technologies for USV swarms usually can guarantee the convergence of the swarm, but consideration on accelerating the formation convergence speed is limited. Most of the time optimal formation control methods focuses on UGV swarms. And optimal control is widely used to solve this problem. However, the optimal control scheme depends on the accurate mathematical model of the system. For USV swarms, it is difficult to establish accurate mathematical model due to the disturbance of water flow, making it inappropriate to adopt this method directly. To solve this problem, a time optimal formation control method is proposed to increase the speed of forming the desired formation. The method is called BA-CPTD formation control method because it is based on Bat Algorithm and Control Parameterization and Time Discretization strategy. As robots suffer from disturbance of sea current, it is difficult to establish an accurate mathematical model. Thus the model of the system is decomposed into a combination of an approximate mathematical model and a compensation term. For the approximate mathematical model, by discretizing the convergence time and parameterizing the change of the control variables into a set of vectors, the formation control problem is modeled as a nonlinear system optimization problem. The Bat Algorithm is adopted to solve this problem, obtaining a control law that can quickly guide the multi-robot system to the desired formation. Thereafter, a Line of Sight based formation control method using real-time feedback is adopted so that robots will eliminate the formation errors caused by the compensaton part and maintain the formation. On this basis, the communication protocol and behavior rules are developed drive robots to monitor the area of interest in formation. Complex behaviors such as formation reorganization and master-slave switching are realized in the patrol process to cope with the detected targets. (3) Regional surveillance method for unmanned surface vehicle swarm based on dynamic forest. This scheme is used to control a USV swarm to patrol in the area in a decentralized manner, using radio communication that can reach long range and exchange large volume of data. When patrolling in formation, the USVs gather together, making it easy to miss the information of other parts of the area. The strategy of decentralized patrol enables the swarm to patrol the whole target water area evenly, detecting the anomalies in each sub area in time and trigger swarm to make further response to the anomalies. So it is an important supplement to the formation patrol strategy. The first difficulty of decentralized patrol is that the patrol area is usually islands, reefs, waterways and other areas. The swarm needs to divide the area with complex border evenly to plan closed patrol path without overlap. Secondly, due to energy supplement, failure loss, task conversion and other reasons, the USVs may join or exit the patrol. The patrol strategy needs to be able to adapt to the dynamic change of the cluster size. And finally, in order to reduce the energy consumption and control difficulty of the unmanned boat, frequent turn should be avoided. These problems are solved by constructing a balanced forest topology that changes dynamically according to the situation in the area of interest. Patrol paths are then planned for each vehicle following the dynamic forest. According to this idea, swarm control methods are developed for two situations, i.e. whether the terrain of the area is known or not. In case that the terrain is known in advance, a centralized control structure is adopted for the swarm. The central node divides the region into several sub-regions according to the shape of the region, and forms the forest by constructing the spanning trees of each sub-region. In order to achieve uniform segmentation of region, the problem is modeled as a problem to homogenize a node weighted graph. An algorithm that can homogenize a node weight graph is proposed, with convergence of the algorithm proved based on Markov process. Based on this algorithm, the uniform segmentation algorithm of the region is obtained. It is further optimized by smoothing the boundary of the sub-regions according to the relationship between adjacent edges. An algorithm is then proposed to plan the patrol path based on the forest, with the behavior rules given to realize self-adaptation to the dynamic changes of the size of the fleet. When the shape of the region is unknown, the contract network mechanism is adoped to guide the fleet to explore the area while planning the patrol path. To achieve this target, two contract network protocols are designed, namely the regional distribution protocol and the regional homogenization protocol. They will modify the spanning tree with the exploration of the area of interest, forming a balanced tree so that proper patrol path can be planned. With this method, the swarm can adapt to dynamic change of the swarm size. (4) Swarm area monitoring method based on virtual pheromone. Behavior laws are designed with bionic ideas for decentralized partrol of underwater robot swarm and decentralized patrol of USV swarm respectively. Most of the existing group patrol strategies are developed by using design scheme, and the coordinated control of the swarm is realized by designing complex algorithms and communication protocols. This kind of methods have high requirements for the computing and communication ability of robots, which limits the application of the method in the marine environment, and also increases the cost of individual robots in the swarm, finally making it hard to a build large-scale robot swarm. The animal community in nature only relies on simple behavior rules and limited information interaction to generate complex swarm behavior and show swarm intelligence. Therefore, the simple algorithm and information interaction strategy can be used to organize robot swarm by mimicking the behavior of social animals. Research for underwater robot swarm and USV swarm is carried out respectively. For large-scale underwater robot swarms, in view of the difficulty and high cost of point-to-point communication between robots, robots are coordinated using virtual pheromone, with the help of underwater communication nodes. The scheme is inspired by the phenomenon that insect swarms communicate indirectly with pheromone. A swarm evolution scheme is adopted to adjust the period for robots to visit the communication nodes. For the USV swarm carrying low-cost communication equipment, aiming at the problem of limited communication bandwidth, the behavior law is developed mimicing the phenomenon that social animals mark their territories with pheromone. Robots will fight for lands according to the density of pheromone. As a result, robots can allocate patrol regions and plan patrol paths will less communicaiton payload.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/25937
Collection海洋信息技术装备中心
Recommended Citation
GB/T 7714
李冠男. 群海洋机器人区域监视方法研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2019.
Files in This Item:
File Name/Size DocType Version Access License
群海洋机器人区域监视方法研究.pdf(8278KB)学位论文 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[李冠男]'s Articles
Baidu academic
Similar articles in Baidu academic
[李冠男]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[李冠男]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.