Çoklu robotlar için harita birleştirme

In the area of mobile robotics Simultaneous Localization and Mapping (SLAM) is a challenging problem. In the literature, there are many solutions to this problem for single robots. However, multi-robot SLAM is a relatively new topic, which has additional issues, such as communication, task sharing and map merging. This thesis takes map merging as its focus and this is examined in terms of the specifications for the unknown initial positions of robots. In the map-merging scenario, every robot localizes itself and generates maps individually and the generated local maps of each robot are shared with other robots. This information sharing can be achieved within different architectures. A distributed approach is used in the study reported in this thesis. This approach does not need a fully connected communication network and a central unit to accumulate the information. This thesis examines the map-merging problem of multi robot SLAM though different approaches in literature. The single robot SLAM problem is solved with Compressed Extended Kalman Filter. The challenging part of map merging is the problem of the unknown initial position is solved with map similarity algorithms, the Delaunay Triangulation and triangle similarity metric. The stochastic search of global transformation matrix is undertaken applying the Random Sample Consensus, which is used for estimation of the transformation between the individual maps created by the robots. In the final step, the overlapping regions of the transferred maps are merged with different algorithms such as Maximum Likelihood, M-Estimator and Covariance Intersection. For experimental purposes, an open code simulator for the multi robot SLAM is implemented. Finally, each algorithm is examined under different scenarios and their performance analyses in relation to simulated and real world datasets, are presented in Chapter 5, which contains the details of the experiments.M.S. - Master of Scienc