Social relation and physical lane aggregator: integrating social and physical features for multimodal motion prediction

Purpose The purpose of this paper aims to model interaction relationship traffic agents for motion prediction, which is critical autonomous driving. It obvious that agents’ trajectories are influenced by physical lane rules and social interactions. Design/methodology/approach In paper, the authors propose relation aggregator multimodal where relations mainly captured with graph convolutional networks self-attention mechanism then fused via mechanism. Findings proposed methods evaluated on Waymo Open Motion Dataset, results show effectiveness two feature aggregation modules trajectory prediction. Originality/value This proposes a new design method extract interactions, attention used in each part fuse different relational features, from other improves accuracy LSTM-based prediction method.