Page - 154 - in Proceedings - OAGM & ARW Joint Workshop 2016 on "Computer Vision and Robotics“

to resolve thedifferent tasks aredescribedandspecificproblemsof homesareaddressed. The paper is organized as follows. Section 2. describes the robot platform and sensor setup config- uration. Section 3. presents our system architecture and implementation overview focusing on the navigation related tasksandcomponents,whichareexplained inmoredetail inSection4.. InSection 5., observed navigation problems are identified and addressed, while Section 6. provides an initial overview of the different navigation functions usage during the trials. Finally, Section 7. includes conclusions,final remarks and futurechallenges. 2. Robot platformandsensor setup The PT2 mobile robot platform (prototype 2) was developed for the Hobbit project by partner Me- tralabs [12]. Two symmetric drive units and one supportig castor wheel constitute the low level locomotionsystem. Asafetyedgebumperprotects theplatformandblocks themotorswhilepressed, preventing the robot from moving while it ishittinganobstacle. The sensor setup is based on two main RGB-D sensors, keeping a configuration similar to the one proposed for a previous prototype of the robot [5, 3]. The bottom camera -used for localization- is placedataheightof35cm. Theheadcamera -usedforobstacleavoidance,userdetection,objectand gestures detection and recognition- is mounted inside the robot’s head, and can be tilted. 2D virtual lasers are created from each of the sensors, considering the largest measurements for localization with the bottom sensor (since they correspond to obstacles further away, like walls) and the closest measurements forobstacleavoidancewith the topsensor. Aheight intervalwithin thewhole3Dpoint clouds is considered for thegenerationof thevirtual scans. 3. Systemarchitecture and implementation Thewholesystemarchitecturehashighmodularity. Tofacilitatedevelopment, codereuse, communi- cationsmanagementandintegration, thepopularRobotOperatingSystem(ROS)[16]frameworkwas used. Metralabs robots run the MIRA (Middleware for Robotic Applications) [4] framework, which manages low level control aspects of the platforms and also provides autonomous navigation func- tionalitiesand theMiracenter tool,which runsacomplete instanceof the frameworkwithagraphical user interface. In order to integrate MIRA into our ROS based system, several interfaces were required. The basic infrastructure of the new interfaces was based on existing interfaces from the STRANDS1 project, modified and extended for our choices and needs. In our case, we decided to use MIRA navigation instead of ROS navigation because it was already well tuned for the current prototype platform and for reasons similar to the ones outlined by the Robot-ERA project team, such as enhanced support and robustness [9]. The required interfaces are implemented in different classes and run as a single ROSnode. Inthefirstplace, thevirtual laserscansgeneratedbyROSnodeshadtoberead,convertedandadapted to be used by MIRA for localization and obstacle avoidance. In the other direction, an interface to provide the current localization pose as a ROS topic was required as well, and the corresponding trasformationsarealso computedandbroadcasted. 1STRANDSproject, EC7thFrameworkProgramme. Grant agreementnum. 600623 154