Scenario2Vector: scenario description language based embeddings for traffic situations

摘要: A popular metric for measuring progress in autonomous driving has been the "miles per intervention". This is nowhere near a sufficient and it does not allow fair comparison between capabilities of two vehicles (AVs). In this paper we propose Scenario2Vector - Scenario Description Language (SDL) based embedding traffic situations that allows us to automatically search similar from large AV data-sets. Our SDL distills situation experienced by an into its canonical components actors, actions, scene. We can then use evaluate similarity different vector space. have also created first kind, Traffic Similarity (TSS) dataset which contains human ranking annotations scenarios. Using TSS data, compare our -with textual caption methods such as Sentence2Vector. find outperforms Sentence2Vector 13% ; promising step towards enabling comparisons among AVs inspecting how they perform situations. hope impact community Word2Vec/Sent2Vec had Natural Processing datasets.