Page - 161 - in Proceedings - OAGM & ARW Joint Workshop 2016 on "Computer Vision and Robotics“

endedupaddingasmallextrapieceof furniture in the longcorridorasanadditional reference. Visual markerswitha fixed posecouldhelpcompletelycorrect localization in thoseareas. When entering a room, it is important that the robot is correctly localized in the transversal direction to the doorway and that the doorway is approached from the front, so doors located at one side of a corridor may cause problems, while doors located at the beginning or end of a corridor are better. In order toapproachdoors fromthefront,avoidinggetting tooclose to thecornersides,ausefulstrategy to try in wide enough places is addingnogoareas at sides of a doorway entrance or at sharp corners (Fig. 6). This way, it is possible to have safer navigation behaviour in wide areas while keeping the capability to go through narrower areas. This provides more flexibility than methods with fixed securitymargins for thewholeoperational area. Figure6. MIRAnogoareascanbeusedtoavoidgettingtooclosetocornersanddoorsides. Notethatgoodenough localization in the transversaldirection is still very important. Nogoareas were also useful to avoid difficult and not allowed areas and rooms in the environments. A few examples are shown in Fig 7. Areas with cables and thin obstacles on the floor and very narrowrooms(usuallykitchens)where thisnon-omnidirectional robotcouldnotmanouverwerealso avoided. It isworthnoting, of course, thatNogoareasareonlyuseful if localization isgoodenough. Other challenging situations were caused by thresholds, bumps on the floor and carpets. In the case of thresholds, we tested commercial and home made ramps when possible (Fig. 8). After testing different configurationsandfindingproper steep limits, the robotwasusuallyable topass thresholds. Inafewcasesproblemswereobservedwithstandardplanningmethods ifanewplancausedtherobot to turn while driving on a ramp. A direct behaviour control may actually improve over plan-based approaches innarrowpassagesandin theseparticularcases,but itneedsseparate implementationand specific, situation-dependent triggering. Regarding localization, itwouldbeuseful to switchbetween 2Dand3Dlocalizationmethods, but again thecomplexity wouldbe increased. Another important thing to take into account is that dealing with the dynamic nature of real envi- ronments is an open research issue to which new projects are dedicating big efforts [19, 1]. Our experience with the previously described approaches showed that minor changes in the position of 161