Search

This paper proposes a novel hybrid solution of SfM (Structure from Motion) and SLAM (Simultaneous Localization And Mapping) for the online generation of a 3D photorealistic map. As it is well known, the SfM can generate a 3D photo map, but it is difficult to get the real-scale as well as to build an online map (i.e., the map cannot be generated on the fly). In contrast, while SLAM frameworks are suitable for online real-scale mapping, they are not adequate for 3D photo map generation. To create a synergy effect, the proposed method combines SfM and SLAM. The way of combination is to use SfM for the generation of local maps and to utilize SLAM for a fusion of local maps in a globally consistent manner. Experimental results show that the proposed hybrid approach enables online 3D photorealistic mapping.

Although there have been many researches on mobile robot localization, it is still difficult to obtain reliable localization performance in a human co-existing real environment. Reliability of localization is highly dependent upon developer's experiences because uncertainty is caused by a variety of reasons. We have developed a range sensor based integrated localization scheme for various indoor service robots. Through the experience, we found out that there are several significant experimental issues. In this paper, we provide useful solutions for following questions which are frequently faced with in practical applications: 1) How to design an observation likelihood model? 2) How to detect the localization failure? 3) How to recover from the localization failure? We present design guidelines of observation likelihood model. Localization failure detection and recovery schemes are presented by focusing on abrupt wheel slippage. Experiments were carried out in a typical office building environment. The proposed scheme to identify the localizer status is useful in practical environments. Moreover, the semi-global localization is a computationally efficient recovery scheme from localization failure. The results of experiments and analysis clearly present the usefulness of proposed solutions.

Although there have been many researches on mobile robot localization, it is still difficult to obtain reliable localization performance in a human co-existing real environment. Reliability of localization is highly dependent upon developer's experiences because uncertainty is caused by a variety of reasons. We have developed a range sensor based integrated localization scheme for various indoor service robots. Through the experience, we found out that there are several significant experimental issues. In this paper, we provide useful solutions for following questions which are frequently faced with in practical applications: 1) How to design an observation likelihood model? 2) How to detect the localization failure? 3) How to recover from the localization failure? We present design guidelines of observation likelihood model. Localization failure detection and recovery schemes are presented by focusing on abrupt wheel slippage. Experiments were carried out in a typical office building environment. The proposed scheme to identify the localizer status is useful in practical environments. Moreover, the semi-global localization is a computationally efficient recovery scheme from localization failure. The results of experiments and analysis clearly present the usefulness of proposed solutions.