In this paper an efficient approach of document layout analysis and reading order determination is proposed for a reading robot. Firstly the input document images are preprocessed to remove noises, connect lines and domains, and to reduce the computation time. Secondly a bottom-up, parameter-independent, two-step layout analysis algorithm based on morphology is used, which outlines the geometry of the maximum homogeneous regions and classifies them into texts, tables, and pictures. Finally the reading order is determined, by a top-down recursive hierarchy algorithm derived from XY-cut, using a set of rules depending on layout information. Important parameters are acquired using statistic information of the given images to adapt to different types of documents. The proposed algorithm is applied to a large number of document images and the experimental results show that it makes the reading robot be able to read paper documents of different languages, even with complex layout structure.