Page - 37 - in Proceedings of the OAGM&ARW Joint Workshop - Vision, Automation and Robotics

a minimal configuration such that the robot can perform a capability. Another method which uses an ontology to describe the environment was presented in [9]. The method uses a de- scription of the environment which is based on an ontology together with a description of skills the robot can perform. Using this description, the robot can plan a sequence of skills which need to be executed to perform a certain task. Such a task was presented in [10] where the robot had to place parts of an industrial kitting operation. To plan which robot can perform which task in a het- erogeneous group of robots the method outlined in [11] can be used. The method defines capabilities which have preconditions which need to be met to allow the execution as well as information which need to be provided by the robotic system to allow the capability to be executed. Thus, the robot can plan which capabilities need to be executed to perform a task. Furthermore, the robot can use the hardware description to check if such an execution is possible. As many capabilities, e.g. grasp an object, may only work under certain restrictions, e.g. size of the object, one can add an approximation description to each capability which defines which conditions need to hold approximately to allow the execution of the capability. This allows to define capabilities in more detail and thus allow a better distribution of tasks among the robotic group. The method outlined in [11] focus on distributing tasks in a group of robots whereas the tool we present in this paper focuses on the configuration of the robotic system during the design phase. Thus, instead of planning which capabilities to use to fulfill a task, we show which different minimal configurations of the robotic system allow the robot to execute the capability. This allows the developer of the robotic system to choose the best fitting set of capabilities. The method presented in [12] extends AutomationML [13] to allow the modeling of a robotic system. This is done by extending the given concepts with robotic specific concepts such as actuators or sensors. Furthermore, the method allows an automatic conversion of AutomationML specifications into an ontology which can be used to check for consistency. This can be used to model a robotic system with Automa- tionML and afterward check through the transformation to an ontology if every dependency is met or if a component is used which does not solve any given task. Beside these checks, further checks can be applied to verify that this system can be realized with the help of ROS [14]. This allows the developer to ensure that the modeled robot can be realized. After these checks are performed one can apply a model-to-text transformation to generate code stubs which ensure proper communication and operation per the modeled system. This allows a faster creation of a robotic system following a model-based approach like the method outlined in [2]. In contrast to our approach, the method does not offer the possibility to check which components are necessary to perform a capability. Thus, the method presented in [12] is alsonotable to specifyaminimalconfigurationwhich fulfills the need for a specific capability as our tool can. Besides the description of tasks for configuration, such a description is also often used to assign a task, e.g. in a multi- robot system. One such example is which uses an ontology to assign tasks is presented in [15]. The method uses an ontology per robot to describe which roles can be performed and how these roles are performed. Additionally, tasks are described in the ontology and how they can be executed through a role which is assigned to a robot. The system uses this ontology to find a matching robot and assigns different roles for different robots to fulfill the specified task. VI. CONCLUSION AND FUTURE WORK The proper configuration of a robotic system for a given task is a time consuming and tedious task. Especially one needs an expert to perform this task to consider all possibil- ities as well as all dependencies. To address this problem, in thispaperwepresenteda tool for theautomaticconfiguration of a robotic system for a given task. The tool uses an ontology-based knowledge base, allowing to reuse publicly available knowledge bases, to describe which dependencies, exist between a task and software and hardware components. Furthermore, we have presented a method to derive a min- imal set of software and hardware components to fulfill a certain task. This allows the user to simply find a possible configuration of the robotic system, that allows the robot to fulfill its task. To allow an easy interaction the tool has a graphical user interface which allows the user to select tasks as well as used components. Thus, the user can specify the currently used components on the robot to check if a new task can be achieved by the robot, or which components need to be added to allow the robot to achieve a given task. Currently, the tool can only be used by a human to decide which components to use to allow the execution of a specific task. It is left for future work to allow the robot itself to use the tool. This would open the possibility that the robot finds alternative solutions to a task during runtime. Thus, the robot could reconfigure itself to react to a fault or changes in its task. Furthermore, currently only a minimal number of components is searched for the configuration, neither computation costs nor investment or development costs are considered in the configuration. It is left for future work to integrate these costs to allow to find a configuration which minimizes the computation effort or to minimize the investment costs. REFERENCES [1] Z. Zhang, “Microsoft kinect sensor and its effect,” IEEE multimedia, vol. 19, no. 2, pp. 4–10, 2012. [2] G. ”Steinbauer and C. Mu¨hlbacher, “Hands off - a holistic model- based approach for long-term autonomy,” in Workshop on AI for Long- Term Autonomy, 2016 IEEE International Conference on Robotics and Automation (ICRA). [3] Knowrob.org. Capability ontology - knowrob. [Online]. Available: http://knowrob.org/kb/srdl2-cap.owl [4] Apache.org. Jena framework - apache. [Online]. Available: https://jena.apache.org/ [5] choco solver.org, “choco-solver.” [Online]. Available: http://www.choco-solver.org/ 37