الفهرس | Only 14 pages are availabe for public view |
Abstract This thesis is concerned with building an intelligent navigation system for mobile robots. The focus is in indoor navigation using map-based approaches. The problems of localiza- tion, path planning and obstacle avoidance are dealt with. Special focus is given to the localization problem using vision. The algorithm used is Monte Carlo Localization (MCL), a probabilistic algorithm which uses particle filters. Some of the enhancements o! this algorithm were studied, namely MCL with random sampling, MCL with planned sampling (also called Mixture-MCL), and MCL with planned sampling and coherence mechanism. These algorithms were implemented, tested and compared. In this thesis, we proposed another algorithm using planned sampling and the idea of coherence. The experimental results showed that this proposed algorithm exhibits better performance than the other implemented algorithms and is more robust to noisy observations than the existing algorithm which uses the mechanism of coherence. An image retrieval system was implemented and used as the observation input for MCL. We proposed an approximation to speed up the calculations of the image retrieval system. We also proposed an idea of how to generate planned samples for MCL from the image retrieval results, and we proposed a way to weight them. A simple representation for the map, suitable for the robot environment in consideration. was proposed. A path planning algorithm, which relies on this map representation, was implemented. A mobile robot was designed, built and used in online experiments and ill collecting evaluation data for offline experiments to perform the comparison study. |