Formalization of driving scenarios is key to define the operational design domain (ODD) of Automated Driving Functions (ADF). Training machine learning (ML) requires huge datasets, that are costly to produce. We propose a toolchain to generate driving scenario video-clip datasets based on the state-of-the-art CarLA driving simulator engine. Scenarios are randomically generated based on a set of parametric features, that are specified by the user. The variability includes both environmental and scenario-specific aspects. As an initial experiment, we have generated a dataset with 200 samples for each one of the 6 implemented classes. The tool is able to achieve a generation rate of about 130 scenarios (7 s. long each) per hour. The tool includes a verification module, which checks the successful completion of each sample.
Developing a Toolchain for Synthetic Driving Scenario Datasets
Cossu M.;Berta R.;Capello A.;De Gloria A.;Lazzaroni L.;Bellotti F.
2023-01-01
Abstract
Formalization of driving scenarios is key to define the operational design domain (ODD) of Automated Driving Functions (ADF). Training machine learning (ML) requires huge datasets, that are costly to produce. We propose a toolchain to generate driving scenario video-clip datasets based on the state-of-the-art CarLA driving simulator engine. Scenarios are randomically generated based on a set of parametric features, that are specified by the user. The variability includes both environmental and scenario-specific aspects. As an initial experiment, we have generated a dataset with 200 samples for each one of the 6 implemented classes. The tool is able to achieve a generation rate of about 130 scenarios (7 s. long each) per hour. The tool includes a verification module, which checks the successful completion of each sample.
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
