Online Learning Methods for Vehicle Safety Applications in Autonomous Driving

<p dir="ltr">The advances in sensor technology and computing capabilities have made it possible to move towards higher degrees of vehicle autonomy. Scenario-based development is an essential part of the validation and development of autonomous driving functionalities. In the realm of scenario-based development and validation for automated vehicles, the categorization of traffic scenarios emerges as a pivotal element. This problem requires an open-world learning approach but most of the machine learning methods used for traffic scenario categorisation work under the closed-world assumption. A closed-world model will classify all the inputs to one of the classes from the training data. An open-world learning method can identify, collect and cluster unknown traffic scenarios and incrementally add new scenario categories to the already existing ones. </p><p dir="ltr">In this work, a hierarchical architecture for the open-world learning method is proposed. The open-world architecture consists of the following components: an openset recognition model, clustering method using known knowledge, class-conditioned generative models and representation learning model. The components in the architecture contain novel machine learning approaches to address the challenging open-world learning tasks, e.g., Extreme Value Theory (EVT) for open-set recognition, Random Forest Activation Patterns (RFAPs) for clustering, class-conditioned generative models for replay, and self-supervised pre-training for representation learning. The proposed architecture is tested using real-world and simulation-based datasets. </p><p dir="ltr">The incorporation of open-world learning into scenario-based development provides a more realistic and comprehensive framework for validating automated vehicles. By acknowledging the possibility of encountering unknown scenarios, the proposed model not only classifies inputs into known but also engages in finding unknowns, clustering and incremental learning. This enhances the system’s capability to identify emerging patterns and seamlessly integrate new traffic scenario classes, thereby facilitating more robust and realistic scenario-based development processes.</p>