Research On Simultaneous Localization And Mapping For Autonomous Vehicle In Dynamic Environments

Simultaneous Localization and Mapping (SLAM) as a key technology for autonomous vehicles to achieve autonomous navigation, is a necessary condition for a truly autonomous vehicle. At present, most SLAM solutions are based on a static environment. But generally, the work environment for autonomous vehicles is dynamic. Therefore, it is necessary and significant to research the autonomous vehicle SLAM problem in dynamic environments.The autonomous vehicle SLAM problem in dynamic environments is mainly related to three aspects:SLAM algorithm, data association, dynamic obstacle detection and disposal. In this thesis, these three aspects are researched and main research contents are as follows:(1) A new autonomous vehicle FastSLAM method which is based on niche particle swarm optimization is presented. Niche Technique and Particle Swarm Optimization are integrated into the classic FastSLAM. Particles diversity and search ability are both enhanced by means of the multi-modal optimization, and then particles are concentrated around the true state of the autonomous vehicle, which can improve the estimation performance of the particle filter and so the SLAM precision is enhanced. The simulation experiment results show that the performance of the improved FastSLAM method is improved significantly, and it can maintain high accuracy in the case of only a few particles.(2) A hybrid approach of data association based on local maps is presented. This approach combines the classic Individual Compatibility Nearest Neighbor (ICNN) algorithm and Joint Compatibility Branch and Bound (JCBB) algorithm. First, ICNN is used for data association in the local map, and the correctness of the result is determined. If the result is incorrect, JCBB is used to correct the result in the local area around mismatched points. The experimental results show that the performance of the proposed method is satisfactory on the speed and accuracy, even in the complex environments.(3) A simple and effective data association method for SLAM in dynamic environment is devised. The static and dynamic features are detected according to spatial and temporal differences, and the associated hypothesis is obtained by the hybrid data association algorithm. According to the outlier nature in the associated hypothesis, the associated dynamic obstacles can be extracted. In addition, the uncertainties are solved by means of two or more detection process of the static and dynamic features. The experimental results show that this method is feasible.The overall framework of the autonomous vehicle SLAM algorithm in dynamic environments is devised by combining the research results of the above three aspects. The simulation and experimental results in outdoor both show that the proposed SLAM algorithm is valid in dynamic environments.