Model-driven Multi-Quality Auto-Tuning of Robotic Applications

For the Simultaneous Localization and Mapping (SLAM) problem, many implementations exist, which meet different requirements w.r.t. non-functional properties (e.g., performance). If those requirements change during runtime, the application should change the SLAM implementation. Implementing the selection of the optimal SLAM-algorithms for robots by hand is time consuming and leads to bad code maintainability by mixing application and adaptation logic. Moreover, the realization of the optimization in code requires the developer to re-implement parts of general purpose optimizers, which impairs reusability. An external adaptation logic selecting the optimal SLAM algorithm addresses the maintainability and programmability issues. A model-driven approach for both, application and selection problem, highly increases reusability. To reach these goals, we propose to use Multi-Quality Auto-Tuning (MQuAT), a model-driven approach to build and operate self-optimizing systems following the Models@run.time paradigm. We evaluate our approach by a case study, where robots have to choose between several variants of a distributed SLAM algorithm.