Seite - 153 - in Proceedings - OAGM & ARW Joint Workshop 2016 on "Computer Vision and Robotics“

Experienceswith RGB-Dbased navigation inreal homerobotic trials∗ P. de la Puente1,M.Bajones1,C.Reuther2,D.Fischinger1,D.Wolf1,M.Vincze1 1 ACINInstituteof Automationand Control. TechnicalUniversityofVienna. Austria. 2 MetraLabsRoboticsGmbHIlmenau. Germany. Abstract Autonomousrobotnavigationisanimportantandchallengingcomponent that isstillmissinginmany real applications. In particular, home environments present open challenges that differ notably from one user apartment to another. Laser sensors cannot perceive objects at all heights commonly found in homes, we investigated the feasibility and suitability of using RGBD sensors for 2D autonomous navigation and a variety of tasks at real user homes. We use the concept of virtual laser scans to integrate RGBD data into mapping and localization methods. For realistic user interaction in actual homes we designed and improved ,over several pilot studies, the robot behavior for tasks such as approaching the user. In this paper, we report the adaptations needed to cope with home-specific challengesusingRGBDsensors as asolution toperceive3Denvironments. 1. Introduction In many areas of application of service robots, mobility plays a role of great importance. Non- industrial real environments present increased complexity and in general are very hard to handle [18,14]. In particular, autonomous navigation in user homes is a challenging aspect of care robotics projects. The SRS (MultiRole Shadow Robotic System for Independent Living) project pointed this out and focusedonthedevelopmentofremotely-controlled, semi-autonomousroboticsolutions[11]. Inother projects, suchasGirafandGiraff++, therobotswerealsoexternally teleoperated[2]andtherewasno autonomous navigation. This missing autonomous mobility has been identified as a key next aspect that needs to be solved. In the Companionable project, autonomous navigation to fixed predefined places was incorporated, but a controlled test home was used [17] instead of different real home environments. RGB-D navigation poses additional difficulties [7, 3, 10]. The reduced field of view, the blind detec- tionareaandthesmallmaximumrangeof thiskindofsensorsprovidevery limited informationabout the robot’s sorroundings. Noisier points, spurious measurements and scale issues in the depth data alsoaffect theperception capabilities. This paper presents our developments, adopted solutions and identified issues to overcome the chal- lenges of home environments using RGB-D sensors. We studied different navigation tasks and the adaptiveapproach to theuserandconducted trials inseveralhomesofolderadults. Thecontributions ∗Thisworkwas partially funded by the EuropeanCommission, projectHOBBIT (FP7-288146). 153