【作者】 雷艳敏；

【导师】 朱齐丹；

【作者基本信息】 哈尔滨工程大学 ， 控制理论与控制工程， 2011， 博士

【摘要】 自从20世纪80年代后期,多机器人系统发展以来,由于多机器人系统具有许多单机器人系统无法比拟的优势,主要体现在：应用领域更广泛、容错性更好、完成复杂任务的成本更低、效率更高、可扩展性更好、并且易于研究和开发等,所以多机器人系统已经成为机器人学研究中一个充满活力、具有良好应用前景的研究方向。在多机器人系统的众多研究问题中,路径规划是其最基本的研究问题之一,是多机器人系统完成任务的根本保证。对于实际应用中的多机器人系统,大多工作在复杂、动态环境下,所以本文结合国家“211工程”建设项目“智能移动机器人系统”,对多机器人系统的动态路径规划方法进行了研究,具有重要的理论意义和实际意义。首先,针对机器人之间存在通信时的多机器人系统的动态路径规划问题进行研究,提出了一种将速度障碍和行为动力学方法相结合的动态路径规划方法。在该方法中,充分考虑动态障碍物和其它机器人的速度信息,为机器人规划了避障和避碰区域,完善了避障和避碰行为的姿态角动力学和速度动力学,可有效的防止机器人在动态路径规划时出现提前避障或躲避不及等避障失败现象。通过仿真,验证了该方法的有效性和可行性。其次,对基于速度障碍的行为动力学路径规划方法进行了两点优化研究：一个是提出利用粒子群优化算法对奔向目标行为、避障行为和避碰行为的三种行为模式进行了融合,从而使机器人能够根据传感器所采集的实时环境信息来获取各个基本行为的权值。并把该方法同基于竞争动力学的行为协调方法进行比较,仿真结果验证了算法的可行性和优越性；另一个是用改进的势场栅格法确定机器人子目标点的位置,并利用子目标点的位置代替最终目标的位置建立奔向目标行为的姿态角动力学,从而使机器人的目标点距离初始点较远的情况下的路径规划更合理,扩大了行为动力学路径规划方法的使用范围。再次,针对没有通信情况下的多机器人系统在未知动态环境下的路径规划问题,提出了一种双层模糊控制器结构,设计了危险度模糊控制器和速度模糊控制器。由于双层模糊控制器的应用,使每一层模糊控制器的输入和输出的模糊关系明确,并且简化了模糊规则的设计。其中,危险度模糊控制器充分考虑了运动障碍物的速度信息,把机器人同障碍物之间的碰撞可能性用基于碰撞时间因子和碰撞距离因子的碰撞危险度来表示,从而更能表达机器人同环境之间的关系,使算法更适合动态的环境；速度模糊控制器的设计充分考虑了目标方位角、障碍物方位角和碰撞危险度的影响,并采用基于行为的思想设计模糊规则,规则中体现了奔向目标行为、躲避障碍物行为和沿着障碍物行走行为,使算法更适合狭长的障碍物等特殊的环境。仿真结果表明该方法是可行的和有效的。最后,针对如何利用神经网络实现模糊规则的自动提取和化简、以及自动确定输入和输出隶属度函数的参数等问题进行了主要研究。提出利用改进的人工鱼群算法进行模糊神经网络的结构优化和参数优化,从而实现了模糊控制规则的自动提取和化简,以及输入和输出隶属度函数参数的自动确定。仿真结果表明,该方法能够减少多机器人系统利用双层模糊控制器进行动态路径规划时的计算复杂性,提高了算法的实时性。另外,通过在人工鱼群算法中对行为评价后的人工鱼增加交叉和变异操作,从而加快了基本人工鱼群算法对最优值的搜索速度和减少陷入局部最优值的可能性,同时提出了适合变定义域的最大步长方法,设计了改进的人工鱼群算法。此外,利用实验室的配备了全景视觉传感器和超声传感器的履带式机器人,对本文提出的动态路径规划方法以及相关技术进行了实验研究,主要有基于全景视觉的人工路标识别实验、机器人的三角定位实验、基于光电编码器的定位实验、行为动力学路径规划方法实验和双层模糊控制器路径规划方法实验等,通过实验验证了算法的可行性。更多还原

【Abstract】 Since the development of multi-robot systems in the late 80s of the 20th century, the research on multi-robot systems has made much progress as it outperforms single-robot system in many respects. The advantages of multi-robot systems mainly include broader applications field, better fault tolerance, lower cost to complete complex tasks, more efficient, more scalable, and easy to study and development. Therefore, the study of multi-robot systems has become a vigorous and prospective research direction in robotics. Path planning is one of the most basic problems in the area of multi-robot systems and it is the fundamental precondition to achieve the task. Because multi-robot systems in practical application mostly work under the complex and dynamic environment, the research on the dynamic path planning has an important academic and practical meaning. Supported by the National "211 Project" construction project "Intelligent Mobile Robot System", this dissertation mainly do some research on dynamic path planning method.Firstly, by combining velocity obstacle and behavior dynamics, a new dynamic path planning method is proposed for MRS when robots can communicate with each other. This method considers the velocity information of the moving obstacles and the other robots, and it also defines obstacle avoidance and collision avoidance region for the robot. In addition, it improves the attitude angle dynamics and speed dynamics of obstacle avoidance and collision avoidance behavior. These unique ideas can effectively avoid failing in dynamic path planning including obstacle avoidance ahead of time or too late. The simulation verifies the effectiveness and feasibility of the method.Secondly, the dynamic path planning method, which combines the velocity obstacle and behavior dynamics, is optimized. In the first place, particle swarm optimization algorithm is adopted to fuse three behavior patterns including move-to-goal behavior, obstacle avoidance behavior and collision avoidance behavior. Thus, the robot could obtain the weight of each basic behavior according to the real-time environmental information collected by sensors. We compare two behavior coordination methods of particle swarm optimization method and competitive dynamics method, the simulation results show the feasibility and superiority of this algorithm. Then an improved potential grid method is utilized to determine the sub-target position of robot, which is used to substitute for the ultimate goal position to establish the attitude angle dynamics of the move-to-goal behavior. This idea makes the path planning more reasonable in case of the robot’s target point far away from the initial point and enlarges serviceable range of behavior dynamics path planning methods.Thirdly, the dynamic path planning problem of multi-robot systems in unknown dynamic environment is considered for the case that communication between robots is unavailable. Double-layer fuzzy controller is proposed to design danger fuzzy controller and speed fuzzy controller. By using double-layer fuzzy controller, the fuzzy relationship of input and output of each layer are clarified. In addition, the fuzzy rules are simplified. The speed information of moving obstacle is fully taken into account by the danger fuzzy controller. The possibility of collisions between robot and obstacles is represented by danger degree which is determined by collision time factor and the collision distance factor. Consequently, it is better to express the relationship of the robot and the environment. The algorithm is more suitable for dynamic environments. The effects of target azimuth, obstacles azimuth and collision danger degree are fully considered when designing the speed fuzzy controller. Fuzzy rules, designed based on the ideology of behavior, reflect the behavior of move to goal, avoiding obstacles and walking along the obstacle. This design principle makes the algorithm more suitable for the special circumstances with narrow barrier. Simulation results show that the method is feasible and effective.Finally, the problems of how to automatically extract and simplify fuzzy rules, and how to automatically compute the input and output parameters of membership functions based on artificial neural networks are considered. So an improved artificial fish swarm algorithm is designed and applied to structure optimization and parameter optimization for fuzzy neural network. Thereby the fuzzy rules can be automatically extracted and simplified, and the input and output parameters of membership functions can be automatically determined. Simulation result shows that when the multi-robot systems carry through dynamic path planning utilize double-layer fuzzy controller, the proposed method can reduce the computation complexity and enhance the real-time performance. Moreover; by increasing cross and mutation operation to artificial fish after the behavior evaluation in artificial fish swam algorithm, the speed of searching optimal value is faster and the possibility of searching result falling into local optimal value is reduced. In addition, the maximum step method is proposed to adapt to variable domain.In addition, experimental test uses the tracked robot which is equipped with panoramic vision sensors and ultrasonic sensors. The proposed dynamic path planning methods and their related technologies are verified by experiments including panoramic vision-based artificial landmark recognition experiment, the robot’s triangle positioning experiment, the odometer-based positioning experiment, the behavioral dynamics path planning experiment and double-layer fuzzy controller experiment.更多还原