Qualitative Spatial Reasoning for Topological Map Learning

Abstract

In this paper we investigate the application of qualitative spatial reasoning methods for learning the topological map of an unknown environment. We develop a topological mapping framework that achieves robustness against ambiguity in the available information by tracking all possible graph hypotheses simultaneously. We then exploit spatial reasoning to reduce the space of possible hypotheses. The considered constraints are qualitative direction information and the assumption that the map is planar. We investigate the effects of absolute and relative direction information using two different spatial calculi and combine the approach with a real mapping system based on Voronoi graphs.     

Why do speakers mix perspectives?

Although considerations of discourse coherence and cognitive processing suggest that communicators should adopt consistent perspectives when describing spatial scenes, in many cases they switch perspectives. Ongoing research examining cognitive costs indicates that these are small and exacted in establishing a mental model of a scene but not in retrieving information from a well-known scene. A perspective entails a point of view, a referent object, and terms of reference. These may change within a perspective, exacting cognitive costs, so that the costs of switching perspective may not be greater than the costs of maintaining the same perspective. Another project investigating perspective choice for self and other demonstrates effects of salience of referent object and ease of terms of reference. Perspective is mixed not just in verbal communications but also in pictorial ones, suggesting that at times, switching perspective is more effective than maintaining a consistent one.     

On Embedding a Qualitative Representation in a Two-Dimensional Plane

Abstract

This paper discusses embedding in a two-dimensional plane a symbolic representation for spatial data using the simple objects, points (P), lines (L), circuits (C), and areas (A). We have proposed PLCA as a new framework for a qualitative spatial reasoning. In a PLCA expression, the entire figure is represented in a form in which all the objects are related. We investigate the conditions for two-dimensional realizability of a PLCA expression, and derive the relation that the numbers of objects in a PLCA expression should have. In this process, we use the well-known Euler's formula. We also give an algorithm for drawing the figure of the PLCA expression that satisfies this condition in a two-dimensional plane and prove its correctness. The algorithm generates a quantitative expression from qualitative expression.     

Geometric problem solving with strings and pins

Humans solve spatial and abstract problems more easily if these can be visualized and/or physically manipulated. We analyze the domain of geometric problem solving from a cognitive perspective and identify several levels of domain abstraction that interact in the problem solving process. We discuss the roles of physical manifestations of spatial configurations, their manipulation, and their perception for understanding problem solving processes. We propose an extension of the classical problem solving repertoire of constructive geometry to approach certain problems more directly than under the compass-and-straightedge paradigm. Specifically, we introduce strings and pins as helpful metaphors for a generalization of the constructive geometry approach. We present classical problems from spatial problem solving to illustrate the ‘strings and pins’ paradigm. Three case studies are discussed: strings-and-pins solutions to (i) the ellipse construction problem; (ii) the shortest path problem; and (iii) the angle trisection problem. Comparisons to formal solutions are drawn. Differences and similarities between the compass-and-straightedge paradigm and the strings-and-pins paradigm are analyzed. Features and limitations of constructive and depictive geometry as well as implications for computational approaches are discussed. The strings-and-pins domain is shown to be more general and less restrictive than the compass-and-straightedge domain.     

顾及星像点分布的恒星相机在轨检校

由于卫星发射前后以及在轨运行过程中,环境因素的变化都可能引起恒星相机参数发生改变,从而导致星敏感器姿态测量精度下降.将多片空间后方交会方法应用于恒星相机的在轨检校.在利用该方法检校时,实验发现检校结果的精度受到参与检校的恒星影像上星像点分布的影响,由此进一步提出了凸包面积百分比准则,该方法自动选取分布较好的影像用于检校,有助于提高检校精度.实验结果证明:基于后方交会进行恒星相机在轨检校时,利用凸包法选取的影像片进行检校的精度明显优于未选片时检校的结果.      

面向稀薄流非线性本构预测的机器学习方法

稀薄非平衡流域内连续介质假设已经失效,主要围绕Boltzmann方程及模型方程对稀薄非平衡流开展理论与计算研究,统一气体动理论格式（UGKS）是其中一种代表性方法。在稀薄非平衡流数值模拟中,Navier-Stokes （N-S）方程连续介质假设已经失效,不能有效描述流场非平衡特征。UGKS方法虽然计算精度高,但速度空间离散导致计算效率低下,多维高速条件下数值计算难以开展。基于数据驱动的思想,在N-S方程与UGKS方法的研究基础上发展出了一种稀薄非平衡流非线性本构关系求解方法（DNCR）。该方法以N-S与UGKS求解器获得的流场数值模拟计算结果作为训练数据集,基于流场特征参数采用极端随机树算法生成机器学习模型,对预测流场中线性黏性应力项与热流项进行非线性修正,并耦合非线性本构关系求解宏观守恒方程得到目标状态稀薄非平衡流动数值解。针对DNCR方法中所采用的机器学习方法-极端随机树模型,通过二维顶盖驱动方腔流算例对高维非线性建模涉及的特征参数选取、参数调优开展了相关验证工作,选取若干典型状态对极端随机树模型的泛化性能开展研究,并评估了相关模型与方法的计算精度与计算效率。     

基于压缩感知的频率和DOA联合估计算法

波达方向（DOA）估计是阵列信号处理的一个重要问题,针对信号的DOA估计问题,本文基于压缩感知理论,提出了一种新的频率和角度联合估计算法。首先利用方向波数的空间稀疏性,建立过完备稀疏字典,然后利用压缩采样阵列结构通过求解最优化问题得到方向波数的高分辨估计,最后利用最优空域滤波实现信号到达角度和频率的配对。相对于传统算法,该算法能够实现多信号DOA的高分辨估计,且经过压缩采样后降低了运算量,仿真验证了该算法的正确性与有效性。     

空间非编码标记特征点中心定位方法与比较

归纳总结了应用于可见光测量的空间标记非编码设计原则,例举了棋盘格型、栅格型、圆型等3种典型的非编码标记图案设计,分别阐述了各标记特征点中心定位方法和计算过程,开展了对比试验,结果表明：圆型非编码标记图案设计及其特征点中心定位精度明显高于其他两种图案,其图案设计简捷、特征点突出易于识别、在像平面中呈现各向同性、点中心定位精度高、计算速度快、噪声抑制能力强。     

A Qualitative Calculus for Three-Dimensional Rotations

We have developed a qualitative calculus for three-dimensional directions and rotations. A direction is characterized in terms of the signs of its components relative to an absolute coordinate system. A rotation is characterized in terms of the signs of the components of the associated 3 × 3 rotation matrix.

A system has been implemented that can solve the following problems: 1. Given the signs of direction and rotation matrix P, find the possible signs of the image of under P. Moreover, for each possible sign vector of · P, generate numerical instantiations of and P that yields that result.

2. Given the signs of rotation matrices P and Q, find the possible signs of the composition P · Q. Moreover, for each possible sign matrix for the composition, generate numerical instantiations of P and Q that yield that result.

We have also proved some related complexity and expressivity results. The satisfiability problem for a qualitative rotation constraint network is NP-complete in two dimensions and NP-hard in three dimensions. In three dimensions, any two directions are distinguishable by a qualitative rotation constraint network.     

基于多台站观测的中国区域地磁指数

利用中国14个地磁台站和全球23个地磁台站的H分量分钟值数据,分析单台站小时幅度指数r_H的时空分布特征,在此基础上结合台站之间r_H指数的相似度度量(残差指数Ra),采用K均值聚类算法将中国14个地磁台站划分为7个区域,根据加权法计算各区域的区域指数Rr.结果表明,r_H指数具有27天太阳自转周变化,季节变化不显著,但仍存在春秋季大而冬夏季小的特征;在空间变化上,r_H随纬度的增高而增大,并且在磁暴期间r_H指数的幅值和形态均表现出明显的经度差异,随地方时呈现晨-昏不对称现象;与Dst指数、SYM-H指数、Kp指数及各区域内台站的H分量观测数据对比分析发现,区域指数Rr能有效反映区域地磁扰动.