Page - 194 - in Proceedings - OAGM & ARW Joint Workshop 2016 on "Computer Vision and Robotics“

RC-racecar, its simulationand themotionmodel. This paper is structured as follows. At first, related research is introduced and the interface as well as the adapted velocity motion model are described. Based on this knowledge, the robot and its simulation are annotated and their basic structure is discussed. Additionally, the trajectories of the two vehicles are compared and the reasons why the trajectories and the motion model deviate from eachotherareexplained. Finally, further improvements for theadaptedmotionmodel inusewith the robot and the simulationare introduced. 2. RelatedWork Autonomousdriving iscurrentlya research topicofbothmajorautomobilemanufacturers likeVolvo, Ford or Nissan and newcomers to the topic of automobiles like Google [6]. At the DARPA urban challenge, universities like the Massachusetts Institute of Technology and the Stanford University present their research accomplishments [3]. An example for research on autonomous vehicles with ackermann drive using ROS is Marvin, the autonomous car by the University of Texas at Austin. Members of the Marvin-Team ported the software of the autonomous car to ROS and shared it this way. The ackermann group represents a community developing open source ROS packages for such vehicles. For the project discussed within this paper, ROS is used because is allows to combine and enhance such packages for navigation and odometry. Twist messages3 and ackermann messages4 are used tocontrol the robot, andodometrymessages5 areusedfor tracking. ThestructureofROSallows tocombineall thesedifferentmessagescontained indifferentpackages into one interface. 3. Interface Theinterface iscreated toensurecompatibilitybetweentherobotsof thefleetandthesimulation. For that reason, the interfaceconverts twistmessages toackermannmessages. It alsoconverts theseROS messages into serial commands and vice versa for those vehicles unable to run ROS. The converting structureof the interface is shown inFigure1. Twist messages are commonly used as motion commands because the six parameters they hold pro- videenoughinformationtodefinemotionsinathreedimensionalspace. Inthefurther, twistmessages holdingonlyone linearvelocityandoneangularvelocityareassumed, since theyprovideenough in- formationformotions ina twodimensionalspace. Ackermannmessagescontainavelocity,asteering angleand information about theacceleration and the jerk. The last twoarenotused for thisproject. Thevelocityof theackermannmessagesequals the linearvelocityof the twistmessages. Thesteering angleof theackermannmessagescanbecalculatedwith theknowledgeof thecarsgeometry. Acurve radius of an imaginary third front wheel is calculated by dividing the rotational velocity of the twist message by the its linear velocity. With this radius, the knowledge of the wheelbase and the usage of trigonometric functions, the steering angleϕ can be calculated. Based on the motion commands the car and its simulation receive, they return odometry messages containing the estimated pose and its uncertainty. This information is calculatedbasedon themotion model. 3Twist Messages: http://wiki.ros.org/geometry msgs (25.04.2016) 4AckermannMessages: http://wiki.ros.org/ackermann msgs (25.04.2016) 5OdometryMessages: http://wiki.ros.org/nav msgs (25.04.2016) 194