MBD and Reconfig. of Mobile Autonomous Systems (MoDReMAS)

Enabling devices to act autonomously requires the existence of knowledge of their capabilities, the environment, possible interactions between the device and the environment, and the tasks to be carried out by the device. Such devices should not only be able to cope with unexpected interactions with the environment but also with internal faults and their effects. Internal faults can always occur, e.g., due to errors in the hardware design, software bugs, or abrasion which leads to broken hardware components. The objective of this project is to provide a methodology and a supporting technology which allows the device to autonomously detect a misbehavior, e.g. an unexpected movement in case the device is a robot, and to identify the root cause of the misbehavior, e.g. a wrong parameter setting. Moreover, the device should re-configure itself in order to either repair the fault, e.g. by changing the parameter, or to overcome the problem, e.g. by degrading functionality but retaining sufficient capabilities to fulfill a pre-specified task. From previous research projects some of the underlying foundations have been gained which we can rely on. However, there are many open problems, and some of them will be tackled in MoDReMAS. This includes coping with dependent faults which often occur in hardware/software systems, and integrating diagnosis results coming from different sources at different levels of granularity. The expected outcome of the project are methods and models which enable autonomous and possibly mobile systems like mobile robots to adapt their behavior after internal faults and unexpected interactions with the environment. Using the results of the project will increase the robustness of systems and help to develop systems which can operate truly autonomously even in inhospitable environments far from home or wherever there is a limited possibility to interact with human operators in order to overcome troubles occurring during a mission.

MoDReMAS focused on providing theoretical foundations as well as algorithms for the challenge of increasing robustness and dependability for mobile and autonomous systems. The far-reaching objective behind MoDreMAS is to enable a robot, or a software agent in general, to react not only on undesired stimuli from users or its environment, but also on faults occurring during operation. This is especially important in cases where maintenance and repair tasks cannot be carried out by external parties, e.g., in space operations, or where such tasks can be avoided without the risk of failing a certain task or mission, saving costs. In order to reach this objective, the goal must be to develop smart adaptive systems. For this purpose, models of these systems and their environments have to be developed that allow to conclude desired behaviour. These models have to enable a system to judge whether a certain mission can still be fulfilled in case of a detected misbehaviour, and how necessary adaptions can be achieved, e.g. via degrading the overall functionality. The contributions of MoDReMAS include handling dependent faults, repair actions, distributed diagnosis and debugging of programs. Dependent faults often occur in reality when one failing component causes another one to fail as well. In many of these cases the replacement of one component will not bring the system into a operational state. Moreover, in case of autonomous systems, the situation is even more difficult, due to limited options for compensating misbehaviour. A decision that does not consider dependencies even might lead to situations that become unhandable. The diagnosis engine developed within MoDReMAS allows for diagnosis systems that consider such dependent faults. When using a fault localization procedure, one is certainly also interested in repairing a system. Repairing a system not necessarily means to put it into full operation. Instead, often enabling a system to fulfill a certain mission is enough. Straightforward, we developed the foundations for repairing a system using diagnosis and functional information in an automated way via an AI planning approach. Once diagnosis and repair was manageable for single systems, we were interested in the distributed case, where (autonomous) systems have to collaborate in order to achieve a given common goal. In this respect, we also contributed to the classification of diagnosis in the distributed case. The last contribution of MoDReMAS is the development of foundations and algorithms that allow for automated program diagnosis and repair. A system that is equipped with such capabilities can in principle repair or adapt itself. In order to make such an approach reliable, testing for distinguishing competing diagnoses, as well as the integration of additional knowledge have to be considered. Again MoDReMAS provides solutions for these challenges.