Systematically Analyzing Synthetic Automotive Data to support Space Filling Curves-based Maneuver Detection

Abstract: Context: In autonomous driving system development, the identification of maneuvers within large datasets is progressively becoming more complex, primarily due to the inherent complexity arising from the multidimensional nature of the data describing these maneuvers. In this context, algorithms based on Space-Filling Curves can be an effective way to index datasets by creating a single-dimensional representation of the maneuvers. If Space-Filling Curves address the identification of maneuvers, there is still the need to produce data that allows their identification. This is typically done using a real vehicle quipped with sensors which poses challenges for the generation of dangerous maneuvers and edge cases. Objectives: This study aims to assess the capacity of using simulated maneuver data to generate Morton code from multidimensional data in order to retrieve similar events from a given dataset and identify edge cases to improve testing and training of Autonomous Driving systems. Methodology: We used the design science research methodology to produce an artifact that applies Space-Filling Curves to simulated data in order to identify edge cases. Conclusion: We show that it is possible to retrieve an existing event from a dataset using Morton indexes generated from simulated data. We describe the potential of Space-Filling Curves to be able to identify edge case maneuvers.