Efficient Vision-Based Navigation of Mobile Robots Based on the LT-Map

Kurzfassung

Long distance navigation of mobile robots is in general based metric maps. Such maps suffer from high memory consumption and poor scalability. Especially, if we consider small mobile robots, efficient memory usage is crutial. A navigation strategy, adopted from insect navigation, generates topological maps. Such a topological navigation is memory efficient and easily scalable, in case of memory shortage. In this work, a navigation algorithm for resource limited systems, which is based on the LTMap, is developed, implemented and evaluated. The LT-Map approach stores landmarks in a tree structure, according to the relation between referring viewframes. The roadmap based navigation system utilizes an omnidirectional camera as input sensor and global image features for landmark identification. The topological algorithm requires no metrical information - only a unit vector towards each landmark is utilized. Different feature tracking algorithms are evaluated and robustification methods are applied. The presented system can generate a map of the current navigated path online and retrace it. Robustness and scalability qualify this approach for long distance navigation of mobile platforms with limited resources. All components and the overall system functionality have been evaluated experimentally and the feasibility of the developed system has been proved.