Abstract
The availability of high-quality datasets is crucial for the development ofbehavior prediction algorithms in autonomous vehicles. This paper highlightsthe need for standardizing the use of certain datasets for motion forecastingresearch to simplify comparative analysis and proposes a set of tools andpractices to achieve this. Drawing on extensive experience and a comprehensivereview of current literature, we summarize our proposals for preprocessing,visualizing, and evaluation in the form of an open-sourced toolbox designed forresearchers working on trajectory prediction problems. The clear specificationof necessary preprocessing steps and evaluation metrics is intended toalleviate development efforts and facilitate the comparison of results acrossdifferent studies. The toolbox is available at:https://github.com/westny/dronalize.