Heterogeneous Trajectory Forecasting via Risk and Scene Graph Learning

Heterogeneous trajectory forecasting is critical for intelligent transportation systems, but it challenging because of the difficulty modeling complex interaction relations among heterogeneous road agents as well their agent-environment constraints. In this work, we propose a risk and scene graph learning method agents, which consists Risk Graph (HRG) Hierarchical Scene (HSG) from aspects agent category movable semantic regions. HRG groups each kind calculates adjacency matrix based on an effective collision metric. HSG driving modeled by inferring relationship between layout aligned grammar. Based formulation, can obtain in situations, comparable performance to other state-of-the-art approaches presented extensive experiments nuScenes, ApolloScape, Argoverse datasets.