A trajectory prediction method based on graph attention network

• 1、Beijing University of Posts and Telecommunications,School of Information and Communication Engineering

Abstract：Trajectory prediction, as an intermediate task in the classic "perception prediction decision" framework, assumes the responsibility of receiving upstream perception information, converting sensor perception information into usable information for utilization; The simultaneously predicted data serves as a reference for downstream decision-making tasks; Therefore, accurate and efficient trajectory prediction is essential, which is also the necessity of trajectory prediction. In order to solve the problem of poor performance of traditional CNN in obtaining global road information in autonomous driving trajectory prediction due to its suitability for processing Euclidean space data, the road structure is modeled as a graph structure, and a Graph Attention Network (GAT) is used for trajectory prediction tasks. The experimental results on the ApolloScape dataset show that the GAT method has a certain degree of improvement in various indicators compared to traditional networks.

Keywords： autonomous driving trajectory prediction graph attention network