Page - 14 - in Proceedings of the OAGM&ARW Joint Workshop - Vision, Automation and Robotics

sensors these sensors uses interfaces which are specific to the type of information they are interested in, e.g. a state change in the component. The information provided by these sensors on the diagnosis port can afterward be used by a monitor. The monitor allows to view the sensing result and thus show which faults are present in the system. This contrasts with the method we propose in this paper as we use the information provided by the observer to calculate a diagnosis. Additionally, our observers allow checking for properties which need to hold between different components, e.g. the movement measured by the wheel encoder and by the IMU. Another method to observe a robotic system was proposed in [12]. Each module in the system is accompanied with a detection module which checks if the module works as expected. This check is performed with the help of a residual calculation. If the residual is not zero a fault is detected. All the detected faults are gathered in a fault signature and used for fault identification. This identification is performed with the help of an incidence matrix. The matrix describes in a static manner which fault causes which observations. This contrasts with our approach as we do not assume that we can simply enumerate all possible observation and faults in a matrix. To react to a fault, the method presented in [12] reacts on the high-level which uses defined recovery actions, which are chosen per the severity of the fault. This is like our approach which use a simple rule engine to perform a reaction but delegates the fault handling to more complex reasoning whereas the rule engine allows a fast reaction. A method which uses a rule system for observations was presented in [13]. The system defines safety rules which are checked during runtime. To define this rules a domain specific languages is used which allows defining conditions for the rules and which actions to trigger if a condition holds. The rules use information which is provided on different topics to define a safety rule. The actions are afterward executed on the robotic hardware and can be defined in the framework separately. The main difference to our system is that we separate the detection and the reaction to a fault. This allows us to use several observations to determine which component is faulty and afterward react depending on the faulty component. As we have briefly outlined above our method is based on the method presented in [3]. The method presented in [3] also uses observers to detect a fault and a diagnosis engine to identify the fault component. Additionally, a planning system is used to repair if a fault is detected. Instead of using a planning system to find a proper repair we use a simple rule engine to allow the robot a fast reaction but also restricts the possible repairs which can be performed. To allow a fast reaction and a proper repair one can combine both methods and first react with the rule engine and afterward trigger a planning step for a proper repair. The other dif- ference between the method presented in this paper and the method presented in [3] is the underlying implementation. The underlying implementation presented in this paper use plugin-based observers which are more efficient than the implementation of the observers used in [3]. VIII. CONCLUSION AND FUTURE WORK Autonomous robots perform tasks in a (partly) unknown environment. This is done by using several complex software and hardware components. These components need to proper function and properly interact with each other to allow the robot to achieve its task. Due to the complexity of the components and the (partly), unknown environment one cannot expect that the robot will perform its task without a fault. Instead one needs to address the problem of fault occurrence in the robotic system. In this paper, we presented a model based approach which allows that the robot detects and identifies a fault. This is achieved by observing the communication between the components and checking this communication for specific properties. These properties are derived from the system and specify the proper function of the system. If a property indicates a fault a diagnosis engine is used to determine the minimal set of components which is faulty. Using the result of this diagnosis engine a simple rule engine can be used to allow the robot to react to a fault. This reaction can be used to repair the fault or to bring the robot in a safe state to perform a more complex repair action. The current approach uses static properties of the system to determine if a fault has occurred. It is left for future work to extend this approach to also consider dynamic changes of the properties. This would allow to detect a malfunction in the dynamic behavior of the system as well as to determine a malfunction of a component which changes its static behavior per a defined system state. REFERENCES [1] 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). [2] M. Quigley, K. Conley, B. Gerkey, J. Faust, T. Foote, J. Leibs, R. Wheeler, and A. Y. Ng, “Ros: an open-source robot operating system,” in ICRA workshop on open source software, vol. 3, no. 3.2, 2009, p. 5. [3] S. Zaman, G. Steinbauer, J. Maurer, P. Lepej, and S. Uran, “An integrated model-based diagnosis and repair architecture for ros-based robot systems,” in Robotics and Automation (ICRA), 2013 IEEE International Conference on, May 2013, pp. 482–489. [4] S. Loigge, “Unified and dependable robot control architecture based on ros,” Master’s thesis, Faculty of Computer Science and Biomedical Engineering, Graz University of Technology, 2016. [5] R. Reiter, “A theory of diagnosis from first principles,” Artificial intelligence, vol. 32, no. 1, pp. 57–95, 1987. [6] A.V.Oppenheim,Discrete-timesignalprocessing. PearsonEducation India, 1999. [7] Student, “The probable error of a mean,” Biometrika, pp. 1–25, 1908. [8] J. Gertler, Fault detection and diagnosis in engineering systems. CRC press, 1998. [9] T. Quartisch and I. Pill, “Pymbd: A library of mbd algorithms and a light-weight evaluation platform.” in 25th International Workshop on Principles of Diagnosis (DX-2014), 2014. [10] C. Mu¨hlbacher, S. Gspandl, M. Reip, and G. Steinbauer, “Improving Dependability of Industrial Transport Robots Using Model-Based Techniques,” in IEEE International Conference on Robotics and Automation (ICRA), 2016. 14