Page - 160 - in Proceedings - OAGM & ARW Joint Workshop 2016 on "Computer Vision and Robotics“

Figure 4. Global path planning in a narrow corridor in a home-like lab. One wall is missing in the map and is not always observed by the obstacles virtual laser (yellow). The initial path may be too close to the wall (left) and only be corrected if the robotgets to face thewall and is notalreadytoo close (right). errors, so a compromise is needed. Fig 5 shows a couple of examples from real user homes. The environmenton the leftpresentsmorecritical localizationconditionswhenentering thecorridors that lookhorizontal in the image, since theyare reachedafter traversinga longfeaturelesscorridor. Along this corridor, however, lateral localization is more accurate because the robot should not accumulate somuchuncertaintybeforeaccessing it fromthese lateral, horizontal, shorter corridors. Figure5. MIRAnogoareascanbeadded inorder toavoidundesiredglobalpathswhenthestaticmapiscleared. In very narrow corridors,however, some marginmust be left toaccount forposible localizationerrors. Today’s laser based localization methods assume that features of the environment can be detected. This is not true in long corridors where the robot could be anywhere along the parallel walls unless an end is observed, which is sometimes not feasible if the corridor is long or when the corridor ends in a room and hence a further opening. This problem can be avoided by using laser sensors, with a significantly longer range. Another option is to combine geometric methods with visual methods [13], but if the walls are uniform the same problem remains. A possible addition of a few external referencesindifficultareassuchascorridors,verywideornarrowspaces,andambiguousplacescould also help in the mapping and subsequent localization processes. In the previous case (Fig 5, left), we 160