Representation Of Uncertain Knowledge And Its Application In Mobile Robot Localization Using Grey Qualitative Method

Generating, storing and representing meaningful knowledge from the obtained uncertain or uncomplete information are prerequisites for intelligent behaviors. By studying and accumulating knowledge, human beings can reduce the uncertainty or uncompleteness of knowledge caused by subjective reasons, such as limits of the cognition capability or capacity of memory by fusing redudant information and supplementing the lacked information by association. Based on the knowledge, standard behaviors with clear goals can be preduced. Mobile robot is one of the typical intelligent agents that have been widely used in both industrial and normal life. More smarter robots like these with abilities of scenario awareness or interaction with human beings have gradually stepped out of labs and appeared in normal life. Due to the compexity and dynamics of the environment, especially limited by resolution of sensors, the information obatained by robots is uncertain or uncomplete, which can be also considered as subjective information from the perspective of a robot. To represent and process uncertain or uncomplete information in a human way, and apply it in mobile robot navigation and localization is of great significance, which makes it one of the hot spots in the research field of artificial intelligence.To process subjective uncertain information, we have to build the sybol and measurement systems to represent and measure such information. Further more, to reason and make decision like human beings, the inferece rules have to be established at the same time. In our research, the grey system theory is studied and based on it we build up the symbol system. At the mean time, the probability theory is used to establish the measurement system. By combing the symbol and measurement systems, the grey probability measure set is built. For conviniece of qualitative reasoning, the traditional qualitative inference theory is adopted and combined with the grey system theory and probability theory to build up the grey qualitative method, which aims to simulate human intelligence in a more natural way. The above mentioned methods are used in the research of mobile robot localization. Simulation and experimental results show that the grey qualitative method is more robust and performs better when there are outliers.The ground work and the main contributions of this work are as below:1) Based on the grey system theory and qualitative theory, the grey qualitative space containing grey qualitative elements and the key points of the elements are built up, which are inspired by the landmark values used in qualitative theory. The traditional whitening fuction used in the grey system theory is also extended to adapt higher dimension applications. Based on the grey qualitative space and the extended whitening fuction, we can represent and process subjective uncertain information without considering too much of its dimension. Further more, the grey qualitative modeling method which is capable of reasoning based on uncertain information and modeling subjective uncertain systems.2) The grey qualitative space and qualitative inference rules are used to simulate the way human beings modeling uncertain systems based on subjective uncertain information. We focus on simulating the process human beings adopt to reason and fuse uncertain information, which means the model is built gradually as the information accumulated to reduce the uncertainty.3) The extended grey qualitative map, which is a novel type of environment model is built up based on the grey qualitative method. While building the grey qualitative map, a environment is partitioned into several levels according to human beings’cognition behaviors. We are no longer consedering higer accuracies as the only goal while processing sensor data, which are ususlly noised. We try to find a balance between accuracy and the capacity of algorithms used to process the uncertain information. The grey qualitative map is used in global localization of a mobile robot in a structured indoor environment. Simulation and experimental results show that the grey qualitative map based localization method is more robust and convenient for establising human-robot interface.4) The state and observation equations are dynamically built, then the grey-dynamic extended Kalman filter is proposed based on the grey qualitative theory and the extended Kalman filter. The algorithm is used to estimate states of the robot and positions of environmental features. To improve the estimation accuracy of the robot heading direction, a map macthing algorithm based on the grey qualitative map is proposed. Both the algorithms are used in pose estimation of a mobile robot and position estimation of environmental features, which show a comparative accuracy and perform well even there are outliers of data obtained by sesors.5) Based on the grey qualitative theory, the grey probability measurement set is proposed, which aims at establishing a representation and measurement system for uncertain information obtained by mobile robots. The system mainly contains two parts:based on the grey qualitative theory, the interval grey number is used to represent uncertain information and the fusion rules of the interval grey number are also given; based on the σ-algebra in probability theory, we build up the measurement sytem of interval grey numbers. The measurement of the fusion process is also given. The grey probability measurement set is then exploit as the basis for a mobile robot localization algorithm. The algorithm is used in the pose estimation process of a mobile robot in unstructured environments and the results show that the algorithm possesses a higher estimation accuracy and can tolerant outliers that arise in the data collection process.