Page - 66 - in Proceedings of the OAGM&ARW Joint Workshop - Vision, Automation and Robotics

thephysicalworld,apermanentsynchronizationof the robots is needed. For this purpose, one of the agents is chosen to act as a leader responsible for collecting and distributing a view of the world and manages reservation of resources. 1) Software Architecture: The function of each robot is separated into the three distinct layers responsible for deliberation (high level), i.e. decision making and planning, a reactive skill engine (mid-level) and low-level components for e.g. motion and vision. The reasoning and planning component is implemented using a CLIPS rule engine [17]. This allows an incremental reasoning to derive at any time-point for each of robot a local optimal decision. The mid-level is designed as a Lua-based behavior engine [18]. With this, simple and complex skills can be modeled as a hybrid state machine. This modularity allows tuning and optimization of skills for specific tasks. The underlying robot framework used is Fawkes [19]. This framework is an alternative to ROS and provides several low- level functionalitiesase.g.AMCL,hardware interfaces to the Robotino base and navigation plugins. 2) Light Detection: The light pattern detection (described in Section IV-A.2) is solved by the Carologistics Team using a more complex and more configuration-intense way. The region of interest (ROI) is cropped using the fusion of the camera and the laser scanner. The robot is aligned with the use of the mounted AR-tag. As this tag can be mounted arbitrarily on the machine, this only allows a course align- ment. With the use of the laser scanner and the knowledge of the type of machine (via the AR-tag), the relative position of the mounted traffic light can be calculated. For this, the exact location of the light for each side of the machine is necessary with respect to the machine base. After this, the region of interest can be restricted a first time. With the knowledge of the position of the laser scanner as well as the camera, it is possible to calculate the position of this ROI in the image frame. Here several heuristics are used to find the shown traffic light, e.g. the fixed width to height ratio, that there have to be three distinct lights stacked in a vertical manner and much more. Having this, the state of the traffic light is determined using the color of the ROIs for the red, orange and green image section. This is done using a defined space for off and on in the YUV color space. Our approach avoids the need for all the configuration by using the HOG-detector and the neural network. The detector eliminates the need for geometric heuristics and knowledge about the machine, and the neural network gen- eralizes enough (trained with several lighting conditions) to detect the state of the traffic light without the need of a tuned color model. This allows more robustness as e.g. different lighting, or a displacement of the mounted camera or the laser scanner would lead to wrong classifications with the solution presented by the Carologistics team. VI. CONCLUSION AND FUTURE WORK With the help of the next industrial revolution, it should be possible to produce individual configured products to the price of current mass-production. This ambitious scheme requires smart factories with modular machinery and an intelligent and flexible transportation system. Such transport can be provided by a fleet of autonomous robots. To offer a standardized testbed for different aspects of such smart factories the RoboCup Logistic League was established. In this paper, we presented a software architecture which can be used to solve various problems appearing in the context of the RoboCup Logistic League. The software ar- chitecture consists of three layers which interact with clearly defined interfaces. The top layer manages the entire robotic fleet,generatesanoptimalglobal schedule, and is responsible for error detection and correction. For this, it uses the mid- layer which provides complex tasks (e.g. explore a zone, deliver a product). Here these skills are decomposed, and the mid-layer commands simple skills (move to a waypoint, open the gripper) to the lowest layer. The software was successfully tested at the RoboCup world championship 2016 and allowed us to rank among the top three teams [20]. Besides the general software architecture, we described in this paper several components in more detail. These com- ponents allow the robot to explore an unknown factory. The presentedcomponents rangefromaschedulingmechanismto distribute the work onto the entire fleet down to mechanisms to detect the type of the machine defined by a signal light pattern. The system is designed in such a way that faults are detected. Thus the system can react to faults properly. This allows the system to reliably to execute its task. To improve the reliability of our system even further the next step is to implement an online diagnosis system as described in [21]. This system can use different measures (e.g. publishing frequency of particular topics, time to respond to actions) to detect abnormal system behavior and furthermore calculate a diagnosis. REFERENCES [1] H. Lasi, P. Fettke, H.-G. Kemper, T. Feld, and M. Hoffmann, “Industry 4.0,” Business & Information Systems Engineering, vol. 6, no. 4, p. 239, 2014. [2] T. Niemueller, D. Ewert, S. Reuter, A. Ferrein, S. Jeschke, and G. Lakemeyer, “Robocup logistics league sponsored by festo: A com- petitive factory automation testbed,” in Automation, Communication and Cybernetics in Science and Engineering 2015/2016. Springer, 2016, pp. 605–618. [3] H. Kitano, M. Asada, Y. Kuniyoshi, I. Noda, and E. Osawa, “Robocup: Therobotworldcup initiative,” inProceedingsof thefirst international conference on Autonomous agents. ACM, 1997, pp. 340–347. [4] U. Karras, D. Pensky, and O. Rojas, “Mobile robotics in education and research of logistics,” in IROS 2011–Workshop on Metrics and Methodologies for Autonomous Robot Teams in Logistics, vol. 72, 2011. [5] F. Zwilling, T. Niemueller, and G. Lakemeyer, “Simulation for the robocup logistics league with real-world environment agency and multi-level abstraction,” in Robot Soccer World Cup. Springer, 2014, pp. 220–232. [6] E. Gat et al., “On three-layer architectures,” Artificial Intelligence and Mobile Robots, vol. 195, p. 210, 1998. [7] M.Georgeff, B. Pell,M.Pollack,M. Tambe, andM.Wooldridge, “The belief-desire-intention model of agency,” in International Workshop on Agent Theories, Architectures, and Languages. Springer, 1998, pp. 1–10. 66