【作者】 杨放琼；

【导师】 谭青；

【作者基本信息】 中南大学 ， 机械电子工程， 2010， 博士

【摘要】 移动机器人定位导航的多传感器信息融合方法一直是机器人研究领域极具挑战性的问题之一,特别是对于未知复杂环境下作业的机器人,如深海底采矿集矿机。它是一台工作在海底的爬行式机器人ROV (Remotely Operated Vehicle),按一定的路线在海底自行移动采集赋存于海底的多金属结核,并通过扬矿系统输送到采矿母船上。当前国内外己作的主要研究普遍集中在通用的移动机器人技术方面,由于涉及技术保密、国际竞争等原因,对于深海底环境下的机器人定位导航等的研究仅仅停留在表面阶段,尚未形成统一和完善的知识体系结构。本研究旨在从前瞻性基础研究入手,基于多传感器信息融合的信息处理方法,针对大洋资源开采与环境勘察的实际要求,开展数据信息处理技术及前瞻信息技术的研发,为移动机器人在海底或水下的应用提供理论基础和设计方法。创新性的研究工作主要有：首先,针对各种类型传感器的性能及误差分析,以移动机器人MKⅡ为实验平台,对定位用内部与外部传感器,如里程仪、红外/激光测距仪、GPS等传感器误差产生原因进行了分析,采用自适应加权融合估计,在总均方误差最小这一最优条件下,根据各传感器所测得的值以自适应的方式寻找个传感器所对应的最优加权系数,使融合后的估计值达到最优。利用该融合算法,实现了GPS与测距法融合的准确导航定位以及红外/激光定位导向的机器人与对接站的精确对接；针对采用长基线系统的水下机器人定位,提出了声速修正的数值迭代模型,并通过仿真分析验证了模型的正确性。其次,以海洋采矿集矿机为研究平台,对集矿机在海底不确定环境下的环境感知、避障及定位导航进行了研究。考虑到深海底环境下的实验较难实现,首先利用移动机器人MKⅡ实验平台,对机器人的状态空间模型、观测模型、融合地图构建、避障策略等问题进行了深入研究,进行了仿真分析与实验验证,并将研究结果应用于深海采矿集矿机的避障、自定位及轨迹跟踪。针对避障问题,由于避障声纳信号很弱,采用SOG (Sum of Gaussians)方法进行传感器的信息融合,再提取特征,使传感器信息的不确定性大大减小。在人工势场法的基础上进行了避障算法的改进,使机器人能够跳出局部极小点,顺利到达目标。对于集矿机的定位导航问题,类比于移动机器人的航程推算法直线模型,建立了履带式集矿机的状态空间模型,根据长基线定位系统的延时建立了系统的测量方程。为了抑制滤波器发散,提高系统鲁棒性,采用基于模糊逻辑控制的自适应卡尔曼滤波(FAKF)的方法及相应的模糊控制规律；由于海底集矿机为履带式结构,其运动模型具有较强的非线性特征的问题,提出了一种改进的SUKF方法以适应系统的非线性。从而实现了基于长基线系统定位的海底集矿机位置估计及轨迹跟踪。最后,进行了未知环境下机器人即时定位与地图构建(SLAM问题)的研究。应用动态阈值的概念,并把动态阈值用于数据相关信息的预处理,即先筛选出符合一定条件的候选观测值集合,以减小后续数据相关操作的计算复杂度,满足SLAM对数据相关实时性要求。在匹配过程中采取基于动态阈值的最近邻数据相关算法,通过分析最近邻数据相关算法在不同置信度下,机器人定位误差的变化情况,以适应不确定性动态变化的实际情况,为提高机器人定位与构图的实时性和准确性提供了保障。在此基础上,针对EKFSLAM算法存在的问题以及机器人运动模型具有较强的非线性,采取基于FastSLAM算法上的Unscented KF方法的措施以改进系统的非线性,实现了移动机器人的位姿估计与环境特征的同时评估。对粒子滤波(PF)基于系列重要性采样算法进行了改进,即基于有效样本大小ESS的重要性权重计算及采样策略。通过在机器人MKⅡ实验及仿真分析,得到了较为满意的结果,验证了算法的正确性。更多还原

【Abstract】 Multi-sensor information fusion for positioning and navigation for mobile robot is one of the most challenging domains, especially for the robot operating under unknown complicate environments, for instance, the miner for sea-bed mining. It is a ROV (Remotely Operated Vehicle) which operates on the sea-bed, collects manganese nodules along a certain route and transports them to mining ship via lifting pipe system.The current main researches are focused on techniques for commonly used mobile robot. For the reasons of technical know-how and international competition, the research on positioning and navigation for ROV under deep sea-bed environments still remains on very low level, the uniform and integrated knowledge system has not been formed. By starting at the prospective basic work and means of multi-sensor fusion process, the research in this thesis aims at processing on data and information, providing theoretical and design methods for the applications of under water or sea-bed mobile robot. The innovative researches are as follows.Firstly, with respect to the characteristics and errors of different kinds of sensors, the errors of proprioceptive and exteroceptive sensors such as dead reckoning, Infrared/laser, GPS are analyzed on the platform of mobile robot MKⅡ. An adaptive weighted fusion estimation is applied, which is on the condition of least mean square errors of GPS and the measures for each sensors, the optimum weighted factors is found to reach optimum estimation for fusion. By using this fusion algorithm, the precision positioning and navigation is achieved by fusing the GPS and odometry signals and the precision docking with an accuracy of less than mms is achieved by infrared and laser guidance docking. With respect to the positioning for ROV in deep-sea mining based on LBL acoustic system, an iterative algorithm model is proposed which is used to correct the acoustic speed. The simulation results show effectiveness of the model.Secondly, on the basis of deep-sea mining platform, the obstacle avoidance and positioning and navigation for ROV in unknown environments are studied. For considering the experiments in the deep sea environments is a little bit difficult to perform, the establishment of state space model and observation model of mobile robot, the building of fusion map and obstacle avoidance tragedy are studied on the platform of robot MKII, simulation analysis and experiments are performed, and then the above researches are applied on obstacle avoidance, self-localization and trajectory tracking for miner in deep sea.As for obstacle avoidance problem, since the sonar signal used for obstacle avoidance is very weak, the information fusion method using SOG(Sum of Gaussians)is introduced, and then the features are extracted in order to reduce the uncertainty of sensor information. Obstacle avoidance algorithm based on artificial potential field method is improved to let the robot skip from the local minimum and get to the goal smoothly. With respect to the positioning and navigation of ROV, The state space model of the crawler type miner was constructed similarly to the dead-reckoning line model of mobile robot. The measurement equation of the LBL system was presented according to the latency of the system. The fuzzy adaptive Kalman filter (FAKF) based on fuzzy logic control and corresponding control rules are applied in order to restrain the divergence and improve the robustness of the system. Since the kinematics model of crawler miner is strongly nonlinear, an improved SUKF algorithm is introduced to meet the nonlinear system. The position estimation and trajectory tracking are achieved based on LBL system.In the end, simultaneous localization and map building (SLAM problem) in unknown environment is studied. The concept of dynamic threshold is applied in the pre-process of data association information. The set of candidate observations are selected to reduce the computer complex and meet the real time requirement. During the match process nearest neighbor method based on dynamic threshold is adopted to suit the change of uncertainty in the real situation and to ensure the accuracy and real-time property for SLAM of mobile robot. On the basis of basic research mentioned above, considering the shortcomings of KFSLAM algorithm and nonlinear characteristics of the kinematics model, the Unscented KF method is introduced to improve the nonlinearity of the system. The position estimation of the robot and environment features are assessed at the same time. The particle filter (PF) based on sequential importance sampling algorithm is improved. The experimental and simulation results show the correctness of the algorithm.更多还原