Page - 156 - in Proceedings - OAGM & ARW Joint Workshop 2016 on "Computer Vision and Robotics“

4. Navigation-related functions Several functions desired by the users required navigation capabilities. This section describes these functions. Moreparticulardetailsofnavigationbetweenfixedplaces in realuserhomesaredescribed inSection5.. 4.1. Gotoplace This is the most basic navigation function. When started by the user, the robot should move from the currentposition toagivenplace,usingpredefinidpositionsandlabels. Theusercanselect thedesired place/roomnamefrom a list displayedon theuser interfaceorcanuseavoicecommand. 4.2. Recharge During the setup phase, the charging station must be placed in a suitable place, which is not always easy to find in real apartments. Enough space for the station itself and its supporting plane that prevents it from moving is needed. There should also be enough space in front of it, so that the robot can detect the station with the bottom RGB-D sensor from a distance (the minimum recommended distance is 50 cm). Obstacles at the sides of the station can also reduce manouverability, increase the risk of false positives in the detection and result in a higher number of failures. Proper wall sockets andsatisfactoryconditions in theroomarerequired,andvisibilityshouldbegood,withnodirect light coming from nearby windows. Last, but not least, it is very important that localization in front of the station is good so that the template is within the field of view (but the error distance to it is not so critical). Therefore, the station should not be placed along a featureless wall with few references in theorthogonaldirection. Places in frontofdoorwaysarepreferredoverpositionswitha lowerdegree ofgeometrical variance in thealignment direction. This task comprises several actions. In the first place, the robot has to reach a predefined position in front of the charging station. From this position, a docking action is started. The docking action starts the MIRA docking procedure, which first of all applies template matching between the bottom virtual scan and the station shape template (recorded from that point during the setup phase). If the template is found, template-based localization is activated and when the robot approaches the station obstacle avoidance is disabled. This procedure was specifically adapted by Metralabs to work with RGB-Dsensors,sinceduringthelastpartof themovementtherobotisblindandopenloopcommands are applied then. When the docking movement is completed, the state machine checks whether the robot is indeed charging or not. If docking succeeded, the robot looks down and the user is notified, otherwisetherobotshoulddockoffandtryagain. If thetemplate isnotfound, theactionisconsidered abortedandthenthestatemachineshouldapplyasmall rotation toonesideandstart theactionagain. If the station template is not detected again, then a rotation towards the other side should be applied and the action is started once more. If detecting the template is still not possible the whole docking task is abortedand theuser is notified. Once the robot detects that it starts recharging (either after autonomous docking or when manually placed into the station), localization is reset to the position of the station recorded and saved during the setupphase. This recoverymechanismwasvery usefulboth for testingandduring the trials. 156