Research On Localization And Mapping For Mobile Robot In Dynamic Indoor Environment

Localization and mapping are two popular topics in the field of robotics. They are regarded as key techniques of building truly autonomous intelligent mobile robots. How to describe the environment and select appropriate state estimation approach to abate system noise are crucial problems for localization and mapping. Based on previous works made by other researchers in this area, this dissertation addresses the problem of localization and mapping in structured environment, especially focused on the adaptability, robustness and efficiency of the algorithm in dynamic environments. The main work includes:1. An innovation map representation named grid-vector is proposed for the mapping problem in dynamic environment. Having inherited the advantages of both the grid map and the vector map, grid-vector map is able to record the uncertainty of en-vironment as well as to reduce storage requirement with the abstract environment information. The dual attribute of grid-vector is perfectly capable of distinguishing dynamic objects from static environmental structures, which is suitable for mapping in dynamic environment.2. An improved localization algorithm based on particle filter is given. It satisfies the demand of long-term operation service robots. The algorithm uses a hierarchical model of the environment which enhances the consistency between the map and the observation model, as well as brings down the disturbances caused by dynamic noise on state estimation. A maximum likelihood estimation method is used to improve the accuracy of importance function to reduce the demand for particle quantity. Using parallel techniques, the improved localization algorithm can be excuted in Graphic Processing Unit, which improves the computational efficiency. As is proved in our experiment, the improved algorithm is able to significantly enhance localization pre-cision and perfectly adapt to dynamic environment. 3. A simultaneous localization and mapping algorithm based on vector matching and particle filter is introduced. The algorithm uses vector map to model the environ-ment, which optimizes storage consumption. It also uses a vector based matching approach to specify sample space and reduces the demand for particle quantity. Be-sides, an adaptive resampling strategy is adopted to reduce the risks of degeneracy phenomenon particle depletion. Comparing to traditional grid map based particle filter approach, this method effectively optimizes execution efficiency and storage consumption. The SLAM algorithm is able to run online in real-time and construct map model in unknow environment accurately.4. A mapping approach is presented based on grid-vector and Expectation Maximiza-tion (EM) for the problem of model optimization in dynamic environment. During the iterations in EM process, the dynamic attribute of objects will be gradually ob-tained and then we can pick up the dynamics information so as to eliminate the influence of dynamic factors in the mapping process. Experiment proves that the dynamic environment mapping approach proposed in this paper is able to perfectly rebuild static environmental structures even in highly dynamic environments.5. A particle filter approach is used to solve the problem of simultaneous localization and map update. This approach uses improved occupancy grid map and discrete-time vector map to record time-varying environment models, so as to estimate the robot pose and map updates simultaneously. The hierarchical model of environment mentioned above ensures the effectiveness of time-varying map update and robot localization in dynamic environment.