生物启发的多机器人系统编队控制与协同搜索

发布时间：2018-01-14 08:31

本文关键词：生物启发的多机器人系统编队控制与协同搜索　出处：《东华大学》2017年博士论文　论文类型：学位论文

【摘要】：多机器人系统(MRS)是多个功能相对简单的机器人通过相互通信、协作等方式组成的整体系统。与传统单机器人相比,其结构松散,冗余度较高,因而具有更好的鲁棒性和并行处理能力。然而由于系统中个体的行为能力有限,所以设计合理的协同控制方式是提升系统整体性能的关键。本论文借助内分泌调节、情感调节以及免疫学习等生物网络调节机理和启发式智能思想,围绕MRS中编队控制和协同搜索两类典型协同问题,针对MRS动态建模、协同编队、协同目标搜索、协同进化等问题展开了一系列深入的研究。本论文旨在深入理解MRS中协同控制机理,从系统的内部状态协同和整体协同规划两个方面构建合理、高效的协同控制算法。本论文的主要贡献如下:(1)MRS编队控制问题广泛存在于工业生产、军事行动及日常生活当中,是系统内部状态控制的基础问题。为了解决系统协同编队和避障等问题,本文基于内分泌调节机理提出了一种新的非线性控制器。通过采用领航者跟随策略,所提的控制算法可以驱动MRS形成任意编队朝目标位置移动,与此同时避免与障碍物发生碰撞。文中理论证明了所提算法在李雅普诺夫意义下是渐近稳定的,且收敛于目标位置。(2)针对领航者跟随策略中固定领航者缺少灵活性和易于陷入局部困境的问题,为MRS提出了一种动态领航者选择模型。其中模糊推理系统被用于根据机器人自身状态和周围环境评估MRS是否被困。在模糊评估的基础上,我们提出了一种情感共识模型以触发领航者重选过程。此外,一种改进型贪婪算法被用于解决在线编队规划这一组合优化问题。通过仿真实验对比不同领航者切换策略,所提的方法可以显著提升机器人编队逃出困境的能力,在领航者故障或被摧毁的情况下,所提模型可以自主地重选领航者,继续完成任务,提高了MRS的鲁棒性和生存能力。(3)除了自身内部的状态协同外,MRS还被广泛应用于目标搜索、灾害救援、目标跟踪等领域,因此需要构建合理的外部运动协同控制算法,以解决强干扰、大尺度的协同搜索问题。本论文以搜救问题为基础,归纳出一类具有非高斯噪声扰动的复杂动态搜索问题。为了解决这一问题,构建了一种基于分布式采样的自适应实时协同搜索模型。模型采用分布式采样和自适应步长控制器,以平衡MRS的探索和发掘能力。采用两种不同信息共享方式,所提模型表现出了不同程度的协同搜索行为。仿真实验表明,在仅采用局部采样的条件下该模型表现出了良好的搜索和跟踪能力。通过分析实验结果,发现所提算法对采样点数量变化并不敏感,这使得该模型可以通过较少的计算量获得较高的搜索性能和跟踪精度。(4)在多目标动态搜索问题的基础上,讨论了在随机可观测条件下对多个逃逸目标进行搜索的问题。结合免疫应答机理,从机器学习的角度提出了一种适用于MRS解决一般动态问题的协同进化网络模型。所提出的模型主要包括两个部分,其中免疫细胞网络负责快速给出复杂动态问题的可行解,免疫学习算法对完整解决策略给出评价,并根据评价不断更新记忆库中的历史策略。亲合度较高的策略将被用于强化训练免疫细胞网络,以提高网络再次求解类似问题时的性能。对所提模型的记忆模块可视化后可以看到,在采用系统重复解决同一动态问题提后,模块中记忆细胞的分布涌现出明显的分簇现象。在搜索过程中系统从随机分配逐渐表现出了协同行为。对比实验表明采取所提模型可以显著提升系统的搜索性能。最后,对全文中所研究的多机器人协同控制问题做了总结,指出这一领域中亟待解决的问题和所采用的方法中的一些不足,并对该领域研究的发展方向和方法做了展望。
[Abstract]:Multi robot system (MRS) is a relatively simple function of the robot through mutual communication, the whole system composed of collaboration mode. Compared with the traditional single robot, its loose structure, high redundancy, so it has better robustness and parallel processing capability. However, because the system of individual behavior ability is limited, so the design is reasonable the cooperative control mode is the key to enhance the overall performance of the system. This paper takes the endocrine regulation, emotion regulation learning and immune network regulatory mechanism and biological heuristic intelligent thought around the MRS formation control and collaborative cooperative search of two typical problems, according to the MRS dynamic modeling of formation cooperative, collaborative target search, launched a series of further study on co evolution problems. This paper aims at understanding the MRS control mechanism of cooperative, collaborative planning two collaborative from the internal state of the system and the whole A reasonable, coordinated control algorithm with high efficiency. The main contributions of this thesis are as follows: (1) MRS formation control problems are widely used in industrial production, military action and everyday life, is a basic problem in the internal state of the control system. In order to solve the system of cooperative formation and obstacle avoidance problem, this paper puts forward the mechanism of endocrine regulation a new nonlinear controller based on. Through the use of leader follower strategy, the proposed control algorithm can drive MRS to form any formation toward the target position, while avoiding collisions with obstacles. The theory proves that the proposed algorithm is asymptotically stable based on Lyapunov functions, and converge to the target position. (2) for leader follower strategy in fixed leader lacks flexibility and is easy to fall into local dilemma problem, put forward a dynamic leader selection model for MRS. The die Fuzzy reasoning system was used according to the robot's own conditions and the surrounding environment to assess whether MRS was trapped. Based on fuzzy evaluation, we propose a model to trigger emotional consensus leader re-election process. In addition, an improved greedy algorithm is used to solve the online formation planning this combinatorial optimization problem. Through simulation experiments different leader switching strategy, the proposed method can significantly improve the ability to escape the plight of the robot formation, the leader of fault or destroyed under the condition that the proposed model can independently complete the task re leader, continue to improve, MRS robustness and survivability. (3) in addition to its own internal state of collaborative. MRS is widely used in target search, disaster rescue, tracking and other fields, so it is necessary to build a reasonable external motion coordination control algorithm, to solve the interference, large scale association The same search problem. This thesis is based on the search and rescue, summed up a complex dynamic non Gauss noise disturbance search problem. In order to solve this problem, constructs a distributed adaptive sampling based real-time collaborative search model. Model with distributed sampling and adaptive step size controller, in order to balance the ability of exploration and MRS explore. Using two different ways of information sharing, the proposed model showed a synergistic search behavior to different extent. Simulation results show that with only local sampling under the condition of the model showed good search and tracking ability. Through the analysis of experimental results, found that the proposed algorithm is not sensitive to the change of the number of sampling points. This makes the model can obtain higher search performance and tracking accuracy with less computation. (4) in the multi-objective dynamic search based on the problems discussed in the machine Observation under the condition of the search problem for multiple target escape. Combining with the mechanism of immune response, from the perspective of machine learning, we propose a co evolution model for MRS network to solve the general dynamic problem. The proposed model includes two parts, one feasible solution of immune cells responsible for network quickly given the complex dynamic problems immune learning algorithm to solve the given strategy, evaluation of integrity, and according to the evaluation of constantly updated memory strategies. The history of high affinity strategies will be used to strengthen the training of cellular network to improve network performance, again to solve similar problems. When seen on the visual memory module of the proposed model can be used in the system. Repeat the same question to solve dynamic, distributed memory cell module in the emergence of clustering phenomenon is obvious. In the process of searching system gradually from random assignment table A collaborative behavior. Experiments show that the proposed model can take significantly improve the search performance of the system. Finally, cooperative control problem for multi robot research in this text is summarized, points out the problems to be solved in this field and the methods used in some of the foot, and the research on this field the development direction and method are put forward.

【学位授予单位】：东华大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP242

【相似文献】