NominalScenes is a dataset geared towards the development of models for dynamic trajectory generation in a navigation scenario. The data includes global plan representations to encode a high level state and plan of an ego-agent and a scene model in order to encode contextual information such as crosswalks, sidewalks, lane markings, and driveable areas. Vehicle speed and relative pose from centimeter-level localization is provided for ground truth labels. Global and scene models are represented in an ego-centric frame of reference.
Paz, D., Zhang, H., and Christensen, H. TridentNet: A Conditional Generative Model for Dynamic Trajectory Generation. In Intelligent Autonomous Systems (Singapore, June 2021) [PDF]