Developing a Testing Environment for Parameter Optimization of Dynamic Movement Primitives

Dynamic Movement Primitives (DMPs) provide a flexible approach for generating goal-directed trajectories. The quality of the trajectories depends on a set of various internal parameters of the DMP. Despite their widespread appli-cation, quantitative guidelines for their parametrization are largely absent in literature. Therefore, the present work intro-duces the methodological foundation for a structured exploration of DMP parameter effects. Building upon the Python codebase dmpbbo for DMPs, which provides the core DMP implementation, an environment consisting of three highly reusable frameworks is developed. It enables systematic parameter exploration: one for small-scale manual trajectory gen-eration, one for large-scale automated trajectory generation, and one for trajectory evaluation based on numerous criteria. Thus, future analyses for determining use case-specific optimal parameters can be carried out with the frameworks. As an application example, a Universal Robots UR5 is considered; in general, the environment can be utilized for trajectory generation in any robotic system. This extended abstract focuses on the design of the software environment, which will enable future systematic parameter studies.