【作者】 杨锦园；

【导师】 黄心汉；

【作者基本信息】 华中科技大学 ， 控制科学与工程， 2010， 博士

【摘要】 由于多机器人系统具备单机器人系统无法比拟的优势,如强鲁棒性、自适应性和高效率。因而,多机器人系统是当前机器人技术领域的一个研究热点,将管理学、社会学、生物学和分布式人工智能学等领域的方法和理论引入机器人学的研究中,从系统的角度探讨多机器人系统的合作行为、组织形式、信息交互和进化机制等问题。本文以多机器人系统为研究背景,结合传感器管理技术,对多机器人系统中信息的融合,多机器人协调中的任务分配与规划等问题进行了深入研究,提供了有效地解决方案,主要工作和贡献如下：本文首先综述了多源信息融合、传感器管理以及多机器人协调控制的研究现状,分析比较了当前该领域的主要研究问题、主要研究方法和关键技术,总结了各自的特点和发展趋势,其中,对外部环境感知和多源信息融合是移动机器人协调控制的基本能力。为此本文介绍了研究中所涉及的基础理论和实验中使用的机器人本体上用于感知外部信息的各种传感器。接着对融合算法进行了两点改进。一方面,DSmT融合算法的效果依赖于广义基本信度分配,即gbba。而广义基本信度分配值的获取都是由专家根据自身的经验提出的,具有很大的主观性。而粗糙集理论仅仅需要传感器的测量数据,而无需其它主观信息,即可归纳出数据间的内在联系。将粗糙集理论与DSmT理论结合起来,根据粗糙集的数据归纳特性,得到gbba的客观获取算法,为下一步推理,即融合过程提供客观依据。另一方面,与DST理论相比,DSmT理论能够很好地解决证据冲突时信息融合的问题,却带来了“焦元爆炸”,致使推理过程的计算量大大增加。同时考虑到DST理论良好的数学基础和冲突情况下的融合效果,本文提出DST-DSmT智能融合算法,力图结合两种理论的优点。在DST融合算法和DSmT融合算法相互转换的过程中,给出了对矛盾焦元的处理方法。该方法充分利用包含冲突的原始信息,目的在于充分反映冲突焦元所提供的信息,以期降低对最终融合结果的影响。为了充分发挥融合系统的功能及性能,需要在环境条件容许的情况下,对有限的传感器资源进行科学合理的分配。首先考虑了传感器在机器人本体上的分布特点,针对不同的信息源在融合过程中的可靠程度以及其对融合结果的影响大小是不同的,提出信息源支持度(MES)概念,充分考虑了焦元之间的相关程度来确定多源系统的信息源的重心,避免了仅仅依据焦元信度分配的数值平均来进行决策的局限性。计算系统中每个信息源与系统信息源重心之间的关系,去除了与当前时刻无关或影响不大的传感器,大大减少了参与融合的传感器数量,从而降低系统计算复杂程度。在此基础上,针对多源信息融合系统中存在大量不确定信息的特点,结合线性规划方法,建立了目标函数、约束条件和优化准则,运用规划论的优化方法得到系统中传感器的管理决策,同时也提高了获得正确融合结果的可能。在上述研究的基础上,对多机器人系统提出动态分区的协作探索策略。设计运动协调混合式的体系结构,给出了行为管理、行为模块和行为决策的功能设计,并设计了4种常用的行为模块。设计了任务协调混合分层的体系结构,并给出了多机器人系统任务协调工作流程。每个机器人根据运动协调机制自主运动,同时任务协调机制也实现了机器人的自主任务分配。在机器人已经完成任务或者不能完成任务时,机器人之间能够根据不确定度变化量的大小自主协商,将任务分配给能最大程度提高效率的的机器人,有效地避免了过多的机器人集中于相同的子任务导致冲突加剧的问题。最后通过在Pioneer2-DXe机器人平台上的仿真获得的实验结果和数据,验证了算法的鲁棒性和可靠性。更多还原

【Abstract】 With a large range of advantages, such as robustness, adaptation and high efficiency, a multi-robot system can outperform a single robot and tend to be accepted in many applications where a single robot system has been thought not be suitable. As a result, the multi-robot system has been paid much attention. Nowadays as a rising subject, the cooperative robotics integrate the theories of management seience, soeiology, biology and distributed AI etc. It discusses many topics systematically, such as cooperative behaviors, arehitecture, communication and evolution of the robot system. Based on multi-sensor management, the dissertation gives a through and systematic research on the information fusion of multi-robot system, the tasks distribution and programming of multi-robot coordination. The main contributions are as follows:Firstly, the paper suvreys the developments of multi-source information, cooperative robotics, sensor management and several influential multi-robot systems. We introduce the main research methods, the characteristics and key technology following the main topics about multi-mobile robots. The paper introduces the involved theories too.Then the two improvements are studied on the fusion algorithm. On the one hand, the effect of DSmT fusion algorithm depends on the general basic belief assignment called gbba. The gbba is obtained by the experience of experts. It’s based on their own knowledge and easy to cause subjectivity and conflict. The Rough Set Theory needs only sensor data without any subjective information to sum up the inter link between the data. Integreting RST theory and DSmT theory, an objective algorithm on gbba has no difficulty to establishied with the summarized character on data of RST. It will provide an objective basis for the further reasoning and fusion. On the other hand, compared with the DST theory, DSmT theory can resolve the conflict evidences in the fusion information successfully. But DSmT theory brought the problem about "explosion of focus elements". It results an increasing calculation in the fusion greatly. Taking into account the excellent mathematics foundation of DST theory and good fusion result with low conflict situations, DST-DSmT intelligent algorithm is presented in the paper to integrate the advantages of both theories. During the conversion process between DST and DSmT fusion algorithm, an approach processing the conflict focus enlements is given. The method takes full advantage of the conflict in the original information to reflect the information provided by the conflict focues enlements fully.It would decrease the impact on the final fusion result.In order to make the best use of multi-source fusion system, the sensor management becomes an important part of data fusion system.Firstly, considering the distribution of sensors on the robot, different information sources have different reliability and impact in the fusion. The concept of Measurement of Evidence Support, MES is proposed in the paper. The correlation of the focus elements is fully taken into account to determine the core of the information sources in the multi-source system. It has avoided the limitation just depending on the average of the general basic belief assignment to make decisions. A distance is obtained according to the relationship between every information source and the core of the sources in the multi-source system. Some sensors which are irrelevant or have little effect in the fusion have been filted to greatly reduce the number of the sensor in the fusion. Therefore the computation complexity is reduced sharply as well. There exists a great deal of uncertain information in the multi-source information fusion system. An objective function, constraints and optimization function are established with linear methods. The decision about the sensors in the system will be obtained with the optimization method. At the same time the possiblity of the correct fusion result will be improved.Based on the above studies, the collaborative exploration strategy for dynamic partitioning is proposed on multi-robot system. A modular hybrid structure for practical application is designed and made applicable theoretic methods for behavior management, behavior process and behavior decision to the structure.A Layered hybrid architecture for multi-task coordiantion is designed. And the task coordinate flow is given to the multi-robot system. Each robot has its own motion occording to the coordination mechanism, and the multi-task coordination mechanisms make the autonomy of the robot task allocation come to reality. When the robot has completed the task, or can not continue its task, the robots can be self-consultation to get the maximum benefits according to the change degreee of the uncertainty. It’s effective to avoid too many robots putting togenther their focus on the same sub-task, which will lead to intensification of the conflicts.At last, we mend the filter process so as to improve the practicability and reliability. Simulations has been done to prove the practicability and reliability of the method.更多还原