SIA OpenIR  > 机器人学研究室
面向废墟内部环境的主动SLAM算法研究
其他题名Study on Active SLAM Algorithm for Ruins Environment
王楠1,2
导师赵明扬 ; 马书根
分类号TP391.41
关键词移动机器人 自主导航 同步定位与地图创建 废墟搜救
索取号TP391.41/W34/2016
页数117页
学位专业机械电子工程
学位名称博士
2016-05-25
学位授予单位中国科学院沈阳自动化研究所
学位授予地点沈阳
作者部门机器人学研究室
摘要本文的研究依托于国家自然科学基金“基于SEM地图的室内非结构环境中的SLAM问题研究”以及国家“863”高技术计划重点项目子课题“废墟洞穴搜救机器人研制”。随机震害下废墟环境特征分布疏密不均、地形崎岖,救援任务具有探测效率等特殊需求,导致SLAM算法无法直接应用于废墟内部搜寻救援。论文系统地研究了SLAM领域里的不确定性信息处理、复杂地图创建、未知环境空间划分、鲁棒性增强以及同步探测规划问题,提出一种基于混合地图的主动层次化同步定位与地图创建方法,以提高算法对复杂未知环境的适应能力和自主探索能力,为移动机器人以后能够在废墟内部环境真正实现自主搜救提供理论性的预研和前期应用性的探索。具体研究内容包括:(1) 基于拓扑米制混合地图的层次化SLAM算法研究针对废墟内部环境所具有的震害形态特殊性,SLAM算法如何综合考虑估计准确度、计算复杂度、环境适用性和交互性的问题,采用层次分析法建立同步定位与地图创建系统地图表示方法构建机制,提出基于拓扑米制混合地图的层次化SLAM算法。采用全局拓扑地图组织整体结构,并在各拓扑节点区域创建局部米制地图,描述破坏形态形成的不规则障碍。通过实验对比,验证了算法环境细节描述和闭环检测能力以及在人工模拟废墟环境的有效性,证明了其在实际搜救和灾情评估等任务中的可行性。(2) 基于谱聚类的大范围未知环境空间划分方法研究震后建筑内部环境受损程度和震害形态分布的无法预见性,导致层次化 SLAM 地图模型转换难以直接预设划分参数。针对该问题,提出一种基于图形分割的区域划分方法,从机器人视角衡量观测区域相关度作为局部子地图创建依据,降低了空间划分对环境先验信息的需求,从而实现层次化 SLAM 地图模型转换。通过对机器人里程和观测信息进行图形映射,基于信息熵生成节点集,将环境相似度作为边的权重,构建无向加权图及相似度矩阵;并采用归一化割策略对图形进行划分,得到以机器人主观方式的环境空间划分结果;方法在解决 SLAM 计算量递增问题的基础上,满足层次化SLAM算法中子地图间的条件独立假设,有效降低了分割造成的信息损失,从而确保了全局一致性。最后,通过仿真及模拟废墟实验,验证算法的有效性和可行性。(3) 基于误差校正和虚拟点修复的SLAM鲁棒性增强方法研究针对感知数据存在歧义的非理想情况下SLAM算法准确度下降的失效问题,提出一种鲁棒性增强方法,从算法改进的角度抑制噪声干扰源的负面影响,实现稳定的未知环境感知和机器人自定位。首先,分析输入数据的误差来源,采用异常检测技术剔除错误数据降低输入量;然后,基于迭代最近点方法对特征提取误差进行校正;最后,提出虚拟还原点针对数据缺失情况修复状态估计量。通过实际环境验证,所提出方法能够有效提高SLAM准确度和容错能力。(4) 基于多目标优化的SLAM主动探测问题研究针对固定探测策略的盲目性导致SLAM算法无法满足搜救效率需求的问题,提出基于多目标优化的探测规划方法,机器人选取最优化目标函数的控制输入实现闭环形式的主动SLAM,从而使机器人以主动、自适应的方式自主规划对未知区域的探测。将SLAM中的探测规划问题转化为多目标优化控制,提出目标函数综合量化评估不确定度、运动代价以及探测增益,并提出基于信息熵的主动闭环约束进行回溯修正。最后,通过对比仿真和实验验证了所提出算法的正确性和有效性。
其他摘要This study is financially supported by the project of “SLAM Based on SEM in Unstructured Indoor Environment”, which is a project of National Natural Science Foundation of China, and the project of “Research of the Rescue Robot in Ruins and Caves”, which is a sub-project of National High Technology Research and Development Program 863. Because of the special requirements, such as the uneven distribution density of environmental feature and rough ground, and the detection efficiency of search and rescue, the SLAM algorithm cannot be directly applied in ruins. In the context of rescue after the earthquake, we study on the SLAM problem systematically from the aspects of uncertainty processing, complex map building, unknown environmental partition, robustness enhancement, and simultaneous exploration. To improve the environmental adaptability and autonomous detectability, an active hierarchical SLAM algorithm based on hybrid map representation is proposed. We aim to provide a basic algorithm that can be used in ruins. The main contents of this thesis are summarized as follows: (1) Hierarchical SLAM algorithm based on hybrid metric topological map To ensure the system computing capability, ambient adaptability and friendly interaction, a map representation method with an effective map building mechanism based on analytic hierarchy process for representation of ruins environments is presented. Based on the hybrid metric-topological map representation, the ruins environment is described at different levels, according to the morphological characteristics in the interior of ruins environments after a seismic disaster. The whole environment is described on the basis of the global topological map, while building a local metric map at each topological node region, to represent clearly the irregular obstacle formed by a destructive pattern. Through the experimental comparison, the ability of environmental details description and closed-loop detection, and the availability at artificial ruins environment are verified. The experiments prove the feasibility of the algorithm at practical tasks such as search and rescue and disaster assessment. (2) Spatial Partition approach based on spectral clustering for a large-scale unknown environment Because of the unpredictability of destructiveness and distribution of seismic damages in the after-earthquake buildings, it is difficult to preset an appropriate parameter of partition for a model transformation of map representation for hierarchical SLAM directly. Based on the graph partition, a spatial segmentation method to achieve the model transformation is proposed. The partition is obtained from the robot's point-of-view by measure the correlation between local areas. Therefore, the requirement of prior environmental information is reduced. The information of odometers and observations of the robot are abstracted as a graph. The generation of nodes is based on the information entropy and the weight of edges is measured by the environmental similarity. An auxiliary weighted graph and the similarity matrix are built. The normalized cut strategy is used to divide the graph. On the premise of solving the problem of computational complexity increment, the loss of relevant information is minimized and the consistency of global mapping is ensured. Finally, the feasibility and validity of the proposed algorithm are verified by simulations and experiments in artificial ruins. (3) Robustness enhanced approach of SLAM algorithm based on error correction In the case of non-ideal conditions with ambiguous perception data, the accuracy of SLAM algorithm is decreased or even failure. From the perspective of improving algorithm to suppress the negative impact from noise source, a robustness enhanced approach of SLAM algorithm is presented to ensure the unknown environmental perception and robot self-localization stably. First, on the basis of analysis of sources of errors in input data, an anomaly detection method is used to reject wrong data and reduce the input quantity. Then, based on the iterative closest point, the error caused by feature extraction is corrected. Finally, a virtual restore point is proposed to repair the estimation of state variables when the input data is unavailability. The validity of the proposed approach is verified by actual environment, the accuracy and fault tolerance of SLAM algorithm is improved. (4) Active exploration of SLAM based on multi-objective optimization The blindness of a fixed exploration strategy cannot satisfy the efficiency of rapid search and rescue requirement. To balance the accuracy of SLAM algorithm and the efficiency of exploration, a subjective exploration strategy in the framework of simultaneous localization and mapping (SLAM) in the form of closed-loop is proposed. By selection of the optimal inputs, the robot can explore unknown areas in an active, adaptive manner. The problem of the subjective exploration of active SLAM is converted into an issue of multi-objective optimization. An objective function is presented to evaluate the uncertainties of estimation, the cost of movement and the gain of exploration. Furthermore, considering the uncertainty of estimation that is measured by the information entropy, an active constraint of loop closure is used for a backtracking correction of accumulated errors. Finally, the feasibility and validity of the proposed algorithm are verified by contrastive simulations and experiments.
语种中文
产权排序1
文献类型学位论文
条目标识符http://ir.sia.cn/handle/173321/19680
专题机器人学研究室
作者单位1.中国科学院沈阳自动化研究所
2.中国科学院大学
推荐引用方式
GB/T 7714
王楠. 面向废墟内部环境的主动SLAM算法研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2016.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
面向废墟内部环境的主动slam算法研究.(7527KB)学位论文 开放获取CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[王楠]的文章
百度学术
百度学术中相似的文章
[王楠]的文章
必应学术
必应学术中相似的文章
[王楠]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。