【作者】 蒋荣欣；

【导师】 陈耀武；

【作者基本信息】 浙江大学 ， 电子信息技术及仪器， 2008， 博士

【摘要】 多机器人学是当前机器人领域中令人激动且富有挑战性的新兴学科，有很强的学科交叉性，涉及到了生物学、管理学、分布式人工智能和控制论等多个领域。多机器人学领域的一个重要组成部分就是多机器人系统协作的研究，从系统的角度探讨机器人群体的合作行为、信息交互及进化机制等。本文研究的多机器人编队导航正是典型的多机器人系统协作课题，该课题被广泛应用于军事、救援、安防和机器人足球等领域。多机器人编队导航是指机器人群体通过传感器感知周边环境和自身状态，协作完成编队，实现在有障碍物的环境中向目标运动。其研究内容包括了协作定位，路径规划，多机器人通信，以及合作编队等关键技术。本文围绕多机器人编队导航的几个关键技术，根据它们在编队导航系统构成中的先后逻辑，展开深入研究。定位问题是编队导航基础而又重要的问题，定位的精度及实时性直接影响到编队及导航的质量。本文针对编队机器人协同定位问题，提出了一种基于多传感器融合的多机器人协同定位方法。该方法以建立一个联合滤波模型为基础，联合滤波模型由三个子滤波器及一个主滤波器组成。首先利用离散卡尔曼滤波器融合机器人内部里程计与陀螺仪，推导出惯导系统状态方程；接着改进了CMVision算法，并结合Adaboost算法训练的分类器及摄像机标定算法获取目标的定位信息，将此定位信息作为观测输入，通过扩展卡尔曼滤波器估计视觉传感器对目标的定位状态；然后激光测距扫描仪结合通过视觉传感器获取到的目标在水平方向的投影跟摄像机光轴的夹角，获取目标的定位信息。将此定位信息作为观测输入，通过扩展卡尔曼滤波器估计激光测距扫描仪对目标的定位状态。最后主滤波器计算各子滤波器的信息分配因子，将各子滤波器估计结果按照信息分配因子权重，利用扩展卡尔曼滤波器进行融合，得到精确的目标定位。在协同定位的基础上，本文针对编队机器人导航过程的避障问题，提出了基于障碍物运动预测的类人方式移动机器人避障方法。该方法使用常速(CV)模型、常加速(CA)模型和当前统计(CS)模型描述障碍物运动状态，利用交互多模型(IMM)算法预测出障碍物下一时间点的位置，速度及加速度。将机器人周围空间由外到内划为规划合理路径区域(PPA)、常规避障区域(NAA)以及紧急逃逸区域(UEA)三个区域。基于障碍物运动状态的预测信息，对进入PPA的障碍物，改进了人工势场法进行局部路径规划；对进入NAA的障碍物，采用变速或绕行策略进行避障；对进入UEA区域的障碍物，采用紧急逃逸算法，计算最佳逃逸角度进行避障。机器人之间的信息交互是多机器人系统研究的基础，也是多机器人编队导航最基本的需求。本文针对采用Ad Hoc组网方式的编队机器人数据通信，提出了一种基于Ad Hoc网络带宽预测方式的实时数据传输模型。发送方在通信过程中，使用跨层方式与接收端反馈方式，周期性提取对传输性能影响较大的几个因素。根据这些因素之间的依赖关系，构建贝叶斯网络预测模型，并将周期性获取的影响因素值作为实时样本数据输入预测模型。预测模型输出下一时间周期Ad Hoc网络带宽的预测值。发送方根据该预测值控制网络发送流量，达到更小丢包率和更大网络带宽利用率的目的。基于上述关键技术的研究，本文针对编队机器人队形变换问题，提出了队形变换最优效率求解模型。将多机器人队形变换模式分为静态变换和动态变换，并确定队列变换能耗(FEC)与队列收敛时间(FCT)作为效率衡量指标。最优FEC效率模型是使得队列中所有机器人移动距离之和最小的极小化模型；最优FCT效率模型是使得队列中移动距离最大的机器人移动的距离最小的极小化极大模型。动态变换的效率模型增加了队形几何中心移动方向与范围的约束条件。通过求取模型的最优解，获取各机器人变换后最优空间位置，并得到最优的队形变换效率。在论文的最后，总结了整个论文的工作，指出了进一步研究和探索的方向。更多还原

【Abstract】 Multi-robot system is an exciting and challenging new field of the robotics;itinvolves biology,management,distributed artificial intelligence and cybernetics.The research of the multi-robot cooperation is an important component of themulti-robot system which studies the cooperative behavior,information interactionand evolutionary mechanism from the system aspect.This thesis focuses on theformation navigation which is a typical subject of the multi-robot cooperation and canbe widely applied to military,succor and soccer robot fields.Multi-robot formation navigation depends on some key technologies which therobot group apperceives through and navigates toward the destination.This researchstudies on these key technologies of cooperative localization,obstacle avoidance,information interaction and cooperative formation.Cooperative localization is an important and basic component of the formationnavigation,whose precision and real-time capability will impact the quality of theformation navigation directly.This thesis proposes a cooperative localization schemebased on multisensor fusion,and a joint filter model is constructed which is composedof three sub-filters and a main filter.First,the motion state function is deduced byfusing the odometer and gyroscope.Then,the target information is extracted from theimage by using the CMVision algorithm and the object detection algorithm which istrained by the Adaboost algorithm with Haar-like features.The localizationinformation of the target will be calculated by using the camera calibration algorithm.The laser scanner obtains the localization information by fusing the angle informationfrom the vision sensor.Lastly,the proposed joint filter model fuses the sensorsub-systems to obtain the precise localization of the target.Based on the cooperative localization,this thesis studies on the obstacleavoidance of the formation robots and proposes an obstacle avoidance scheme basedon the obstacle motion prediction.The motion state of the moving obstacle isdescribed using the Constant Velocity (CV) model,the Constant Accelerate (CA)model and the Current Statistical (CS) model.A Kalman-based Interacting MultipleModel (IMM) filter is adopted to estimate the obstacle motion trend.This avoidancefield is divided into three areas:Path Planning Area (PPA),Normal Avoidance Area(NAA) and Urgent Escape Step (UES).An improved Artificial Potential Field (APF), a normal avoidance strategy and an escaped algorithm are designed to avoid anyobstacles who invade into the relevant area.Information interaction is another key technology of the multi-robot system.Themulti-robot requests different throughput and real-time capability when implementingdifferent tasks.This thesis proposes a real-time data transmission model over Ad HocNetwork based on the bandwidth prediction.This transmission model is composed ofCross-Layer,Feedback and Bayesian network techniques.The sender extracts theimpact factors of the data transmission by using Cross-Layer method and Feedbackmethod.According to those relevant factors,this thesis constructs a Bayesian networkprediction model and predicts the bandwidth of the next period.The predictedbandwidth would be then used for the flow control of the sender.Based on the aforementioned key technologies,an optimal efficiency model isproposed for the multi-robot formation transform.The formation transform is dividedinto a static transform mode and a dynamic transform mode,and the formation energyconsumption (FEC) and the formation convergence time (FCT) are adopted toevaluate the efficiency of the formation transform.The optimal FEC model is aminimization model which minimizes the displaced distance sum of every robot.Theoptimal FCT model is a max-min model which minimizes the maximal displaceddistance of the robot.The efficiency model of the dynamic transform is subjected tothe constraint that the geometry center of the formation must move forward thepositive direction and a fixed range.The optimal space position and formationtransform efficiency are then obtained by solving the efficiency model.Lastly,we summarize the general work of this thesis and propose a short outlookon possible future research.更多还原