SceNDD++: An Enhanced Scenario-based Naturalistic Driving Dataset

This paper presents SceNDD++, a novel scenariobased naturalistic driving dataset and an enhanced version of the previously published SceNDD dataset. Naturalistic driving data collected in different regions in the USA, including Austin, San Diego, Washington DC, and Indianapolis was included in the new dataset. The dataset encapsulates a wide array of real driving scenarios with heterogeneous trajectories and behaviors. It comprises data from both the ego vehicle and non-ego actors, incorporating elements such as waypoints, velocity, yaw angle, and dimensions of the bounding boxes. The dataset allows users to incorporate additional actors, sensors, roads, and obstacles. Compared to its predecessor SceNDD, SceNDD++ is more robust by integrating different vulnerable road users (VRUs), such as pedestrians and bicyclists, and featuring smoother trajectories of non-ego actors.