AI-Native Edge Intelligence for Mobility Networks: A Case Study on Attention-Augmented Vehicle Trajectory Prediction

Abstract

AI-native edge intelligence is emerging as a key enabler for next generation mobility networks, where efficient trajectory data transmission and compression must jointly satisfy high accuracy, reliability and data efficiency. However, real time trajectory prediction in dynamic vehicular environments requires the rapid processing of large volumes of heterogeneous, noisy, and redundant data under constrained computing resources. As a representative case study, this paper addresses dynamic vehicle trajectory prediction in vehicular networks. To address these challenges, we propose an edge-oriented framework that integrates self-attention with trajectory feature embedding, state updating, data compression, and trajectory decoding. Within this framework, the self-attention mechanism adaptively identifies salient spatiotemporal dependencies among the target vehicle, surrounding vehicles, roadside sensing information, and traffic flow conditions. This framework effectively captures vehicle movement patterns and dynamically predicts future trajectories. Finally, simulation results validate the effectiveness of the framework in achieving accurate trajectory prediction and efficient processing of edge-generated mobility data.