A mobile robot self-localization approach based on unidirectional vision

This paper presents a self-localization system for mobile robot in large-scale indoor environments. For a structured corridor environment, the vision information is adopted to track the robot pose with a predefined hybrid metric-topological map. A nonlinear unidirectional camera model is developed to project the probabilistic map elements with uncertainty manipulation. Extended Information filters are deployed to estimate the robot pose. The proposed system can perform localization tasks on-the-fly, with the features of efficient map modeling and computational simplicity. Experimental results are provided to demonstrate the performance and effectiveness of the proposed techniques.