The thesis does researches on multi-sensor data fusion, focusing on a specific application - autonomous mobile robotics. Multi-sensor system is one of the most important parts of an intelligent mobile robot. The goal of this thesis is to develop a high-performance, low-cost embedded multi-sensor data acquisition system, construct and maintain a map of the environment using multi-sensor data fusion algorithm on the basis of the system, and improve the ratio of performance to cost of it and the whole mobile robot. Main work and contribution of the thesis is given as follows: 1. After investigating various multi-sensor systems used in mobile robots currently, we present a concept of "embedded environment perception system". Based on it, a multi-sensor data acquisition system for mobile robots is developed successfully, which can collect data from multiple sonar sensors, infrared sensors, PSD and orientation sensor, etc. Experimental results on a real mobile robot show its reliability and real-time capability. 2. Aiming at applying sonar sensors to mobile robots, we analyze their operation principles and performance characteristics, and explain the limitation when sonar sensors are used in mobile robots. At the same time, different mathematic models are introduced. 3. Taking into account of the uncertainty of sonar sensors, we discuss the task of map-building with mobile robots. A new sonar model is given, which is a simplified Gaussian model, to construct a two-dimensional histogram grid map. Furthermore, RIC (Range-Incidence Combination) method is introduced, which fuses incidence angle information and range measurement of sonar to build map of environment. The method exhibits a good performance in solving the problem of door-finding and reducing phantom targets. 4. To overcome the limitation of the domain of application when using histogram grid map (only useful in obstacle avoidance), we propose an alternative to represent the environment-evidence grid map. Dempster-Shafer Evidence Theory is applied in map-building. Moreover, to reduce phantom targets, we fuse data from sonar sensors and long distance measuring sensors, and define Evidence Conflict Value. Comparative experimental results show that the new method can improve the performance of sonar sensors in map-building.
修改评论