Research On Simultaneous Localization And Mapping Of Mobile Robot In Dynamic Environments

Simultaneous Localization and Mapping (SLAM) is a fundamental and hot problem in mobile robotics. It is one of the most important qualifications for mobile robot to realize autonomy. SLAM is the process that a robot locates itself with sensing and estimation, while incrementally builds the environment map. Now, most of the SLAM methods are based on the assumption that the environments are static. Since the real world are dynamic, it is necessary to research the mobile robot SLAM approaches in dynamic environments.There are three sub-problems involved in mobile robot SLAM in dynamic environments: SLAM solutions, data association and dynamic objects disposal. Dedicated to the three sub-problems the thesis develops a deep research, and finally constructs and realizes a SLAM system in dynamic environment which is implemented successfully in the mobile robot MORCS-I. The main contributions of the thesis in research are shown as following:An approach of Simultaneous Localization and Mapping based on local maps is presented. A local frame of reference will be established periodically at the position of the robot, and the local map is then fused into the global map according to the range of the local map to update the global map estimate. This approach to the SLAM problem allows the less computation complexity because of not updating global map each time, and because of the independence of the local maps the approach will not cumulate the estimation and calculation errors.A particle swarm optimized simultaneous localization and mapping method is presented by introducing particle swarm optimization into the SLAM process, aimed at the problem of needing a large sample size. The method exerts the advantages of the particle filter which is applicable to any non-linear system. By considering the effection of the individual and sworm particles, the method can also improve the efficiency of the sample process, as well as reduce the particle number while guarantee the accuracy of the SLAM and the particle's convergence at the same time. In order to resolve the problem that the classic ICNN and JCBB method can not modify the previous association hypothesis, the data association problem is transformed as the discrete optimization and a new multiple hypotheses data association method based on the particle filter is presented. Multiple particles are used to maintain the multiple data association hypotheses, and every particle's weight is calculated by association cost. During the resample, the wrong hypotheses are discarded through basic Branch and Bound approach. The real optimization or suboptimization association results are obtained after a period. By experiment results analysis and comparison, the new method can reach more correct data association results and higher location presicion.An approach for mobile robots is suggested in this paper to map dynamic environments using sonar and vision sensors, aiming at the problem of the low mapping accuracy caused by the fact that the sonar can not detect dynamic objects correctly. In order to maintain more complete mapping information, the approach builds static and dynamic grid maps respectively. The two types of occupancy grid maps are updated by the previous static maps and the sonar observations and the detected dynamic objects using monocular camera. The map update model is given. The vision information can get rid of the mistakes that some dynamic objects are considered as static ones and help build the map correctly.A SLAMiDE (SLAM in Dynamic Environments) system is designed and realized in the thesis, which supplys a holistic structure and a series of implementation methods for mobile robot SLAM in dynamic environments. In SLAMiDE system, the observations are k-neighbour clustered firstly, and a uniform target model is constructed. The dynamic objects and still objects and the mobile robot pose are estimated simultaneously, by synthesized the above research about data association and dynamic objects detection and SLAM solution and local maps. Finally, the results of the experimental test prove that the SLAMiDE system can realize dynamic objects detection and mapping and location correctly.