Orientation finding using a grid based visual compass

In this paper an extension of the model-based visual compass is presented, which can be updated continuously, allowing a robot to orient itself in a changing environment. To build a model, colors in the image are discretized to an automatically generated color profile, and transitions between these classes within vertical lines are used as feature. Experiments show how well a model can be learned at the center of field and how this model can be extended to other location with a randomly walking robot. Finally, the strength of the approach is demonstrated in a dynamic environment, where a good estimate of the orientation is maintained while the surroundings are changed in a controlled way.