Page - 9 - in Proceedings of the OAGM&ARW Joint Workshop - Vision, Automation and Robotics

A Model-Based Fault Detection, Diagnosis and Repair for Autonomous Robotics systems Stefan Loigge1 and Clemens Mu¨hlbacher1 and Gerald Steinbauer1 and Stefan Gspandl2 and Michael Reip2 Abstract—Autonomous robots comprise of several complex software and hardware components which interact with the en- vironment to fulfill a certain task. Due to the non-determinism, inherent of the environment and complexity of the components one cannot expect that the robot will never show a fault. Instead one needs to deal with the occurrence of faults in the robotics system. As we focus on autonomous robots the robot should deal with faults in an automated fashion. In this paper, we present a model-based fault detection and diagnosis method with a simple but powerful method to repair faults. Using this method, the robot can detect and react to faults in a timely manner. Furthermore, no human intervention is necessary thus allowing the robot to be autonomous. As not every repair can be performed by the robot itself the system allows the robot also to inform the maintenance staff which repairs are necessary. Thus, this approach reduces the time for fault localization of the maintenance staff. I. INTRODUCTION Autonomous robots perform tasks in (partly) open envi- ronments. To perform such a task, the robot uses several complex software and hardware components which interact with each other. Due to the (partly) open environment and the complex components, one cannot assume that no fault will occur. Instead one needs to design the robotic system with faults in mind. Thus, one either add fault handling in each component or one uses a more general approach. One such general approach is the use of a model-based approach as outlined in [1]. The model is used to describe the system behavior and to allow the system to detect a fault. The use of a model-based approach allows the robot to determine if a fault has occurred. Furthermore, the robot can determine which component most likely caused this fault. Using the information which component is faulty the robot can determine which action to perform to react to this fault. Besides the possibility that the robot detects and reacts to a fault a model-based approach also allows to separate the current system description from the fault detection and localization components. As the model is used to describe the system the fault detection and localization can be done on the model only. Thus, one can use the software to perform this reasoning for many different robots without changes. The only thing which needs to be changed for a robot is the model of the robot. As many robotic system 1Stefan Loigge, Clemens Mu¨hlbacher and Gerald Steinbauer are with the Institute for Software Tech- nology, Graz University of Technology, Graz, Austria. {sloigge,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 reuse components of other robots, or have similar robot components one can often reuse parts of already existing models. Thus, further decreasing the effort to perform fault detection and localization. In this paper, we present such a model-based diagnosis approach. The method uses several different observers to ob- serve properties of the system. These properties are observed to detect a fault. With the help of the observed properties, the system can derive a diagnosis which component caused the fault. This allows pinpointing the fault without extra costs as the only information necessary for the diagnosis is already provided through the definition of the observations. To allow the robot to react to a detected fault a simple rule engine can be used. The rule engine allows the robot to react fast to a fault and to trigger more complex repair actions. Through this fast reaction, one can reduce the chance that a robot will endanger itself or pose a threat to its surrounding. The remainder of the paper is organized as follows. In the next section, we will give an overview of the fault detection, diagnosis, and repair system. The proceeding section dis- cusses the different observers which check system properties in more detail. Afterward, we discuss the diagnosis engine which is used to identify the faulty component. In Section V we discuss the rule engine and how it can be used to react to faults. In Section VI, we show a use case where the system was used on an industrial robotics system. Before we conclude the paper, we discuss some related research. Finally, we conclude the paper and point out some future work. II. SYSTEM OVERVIEW To create a robotic system, the robot operating system (ROS) [2] is often used as a framework. With the help of ROS one can use several software components, which are called nodes, and interact with each other. This interaction can be performed with the help of publisher-subscriber principle which allows exchanging message between each ROS node. To define and identify for such communication channel ROS uses so-called topics. These are strings defining an n-to-n communication channel. Furthermore, one can use service calls to provide a service from one component to another. In the remainder of the paper, we will focus on messages exchanged by topics as these are used more often as services and allow an easy introspection. Using ROS, a robotic system can be created which uses several software components interacting with each other. As we are interested in detecting and identifying faults and react to these faults we use the system depicted in Figure 1. The 9