Active learning of driving scenario trajectories

Annotated driving scenario trajectories are crucial for verification and validation of autonomous vehicles. However, annotation such based only on explicit rules (i.e. knowledge-based methods) may be prone to errors, as false positive/negative classification scenarios that lie the border two classes, missing unknown or even failing detect anomalies. On other hand, labels by annotators is not cost-efficient. For this purpose, active learning (AL) could potentially improve procedure including an annotator/expert in efficient way. In study, we develop a generic framework annotate trajectory time series data. We first compute embedding into latent space order extract temporal nature Given embedding, becomes task agnostic since can performed using any method query strategy, regardless structure original Furthermore, utilize our discover trajectories. This will ensure previously types effectively detected included labeled dataset. evaluate proposed different settings novel real-world datasets consisting collected Volvo Cars Corporation. observe constitutes effective tool labeling well detecting classes. Expectedly, quality plays important role success framework.