Page - 32 - in Proceedings of the OAGM&ARW Joint Workshop - Vision, Automation and Robotics

Task-Dependent Configuration of Robotics Systems Alexander Pagonis1 and Clemens Mu¨hlbacher1 and Gerald Steinbauer1 and Stefan Gspandl2 and Micheal Reip2 Abstract—To solve a task, a robotics system uses several different hardware and software components. Each of these components solves a specific subtask to allow the overall task to be solved. Thus, the proper selection of the set of components is crucial for the success of performing a task. This selection can become complex if one needs to consider that each of these components has its own dependencies which need to be fulfilled to work properly. Due to this complexity, the proper selection of components is time-consuming and error prone. Additionally, domain knowledge is necessary to consider all dependencies correctly. To proper choose the components without the need of a domain expert one can follow a model based approach. In this paper, we show how such a model-based approach can be used. We present a tool that, based on a domain model, automatizes the selection of the necessary components to implement a set of given tasks. Due to this automatic selection mechanism, one can either simply check if a robotic system can perform a task or which components need to be added to allow the robot to perform the given task. I. INTRODUCTION A robotics system consists of several hardware and soft- ware components which interact with each other to achieve a given task. The selection of the hardware and software components is often done by a domain expert, ensuring that the task can be fulfilled with the given selection. This is a time-consuming task, as one needs to know which dependency each component has, e.g. a computer vision algorithm depends on a camera but does not specify which camera exactly. Additionally, one possible needs to consider many possibilities how a dependency can be met to find an optimal selection. Following simple scenario is used to highlight these difficulties: The task the robot must fulfill is to localize itself. One could now use a localization which is based on a laser or a localization which is based on the camera. In case there is a Kinect camera [1] available but no laser, a camera-based localization approach would probably be preferred. But one could use the depth image to simulate a laser scanner and thus use also the localization based on a laser scanner. This simple example already shows that one needs to consider several possibilities and necessary dependencies to allow a robot to solve a task. 1Alexander Pagonis, Clemens Mu¨hlbacher and Ger- ald Steinbauer are with the Institute for Software Tech- nology, Graz University of Technology, Graz, Austria. {apagonis,cmuehlba,steinbauer}@ist.tugraz.at This work is partly supported by the Austrian Research Promotion Agency (FFG) under grant 843468. 2Stephan Gspandl and Michael Reip are with incubedIT, Hart bei Graz, Austria. {gspandl,reip}@incubedit.com Instead of choosing the hardware and software compo- nents manually, one can follow a model-based approach for the robotic system as it was outlined in [2]. The idea is to use a model that describes the task as well as the available hardware and software components, their capabilities and dependencies. By using this model one can automatically generate a list of components that are necessary to fulfill a task. The model does not only allow to generate a list of components to fulfill a task but it also allows the robot to check if a task can be executed with the given hardware and software. Furthermore, the robot can use the model to decide which alternative software and hardware modules to use if one part of the system does not work correctly. Such a reconfiguration isof special interest if oneconsiderscomplex tasks which can be achieved through several means. In this paper, we present a tool which allows perform- ing such a model-based configuration of a robotic system automatically. The tool can be used to derive which set of components needs to be present to allow fulfilling a task. Furthermore, the tool allows checking if a given robotic configuration can fulfill a task. Additionally, all possible component compositions that allow solving the given task can be viewed. This allows checking which alternatives are possible and which components are redundant in the system. To allow an easy configuration the tool does not only suggests possible configurations but also allows to interactively vary the given configuration. This makes the configuration process easy and allows for a quick decision on the best fitting set of components. The remainder of the paper is organized as follows. In the next section, we discuss the design of the configuration tool. This description comprises the used knowledge base, the method which is used to derive a correct configuration, and the description of the user interface. The proceeding section discusses a simple example scenario and presents how the tool can be used. This is followed by a section discussing the limitations of the approach. Afterward, we will discuss some related research. Finally, we conclude the paper and point out some future work. II. THE CONFIGURATION TOOL As we motivate above using a model one can automate the generationofaconfigurationforagiven task.Thisgeneration uses the model to determine the dependencies between software component and hardware component. Furthermore, the model describes the different possibilities to resolve a dependency. To ensure that the model can answer a query in a timely manner and to allow still the model to be expressive 32