Autonomous Mobile Robot Intrinsic Navigation Based on Visual Topological Map

It is a necessity for mobile robots to navigate in indoor environments. With the navigation ability, robots can move around automatically. Navigation is also the basis of many kinds of tasks, such as object search, home surveillance and human guiding. Nowadays, navigation mostly relies on Simultaneous Localization and Mapping (SLAM). The metric map created by SLAM can provide rich details for navigation. Nevertheless, the drawback of this kind of map is that it is computationally expensive, disk space consuming and lacks of semantic meanings. Therefore, we propose a novel topological mapping and navigation method, which uses image comparison to imitate the navigation process of human beings. We use neural network to do image comparison and propose a image based localization mechanism. With this mechanism, robot can judge whether it has reached the goal position with more flexibility compared to traditional grid map method. This resolves the problem of the lack of semantic meaning of traditional grid maps. Furthermore, the map size of our proposed topological map is tiny. We only need small disk storage space to use our topological map. In summary, our proposed topological map is efficient, disk storage saving and has more flexibility with the aid of vision. Our proposed method has conquered the disadvantages of traditional metric map based navigation systems.