LITNT: A Target-Driven Trajectory Prediction Framework with Lane Change Intent Analysis

Predicting the trajectories of surrounding vehicles is crucial for autonomous driving. In view of the lane change intentions and future states of moving vehicles in complex traffic scenarios, a new target-driven trajectory prediction model that integrates lane change intention prediction was proposed. The graph neural networks were employed to model the interactions between high-definition maps and trajectory data. Subsequently, a lane segment-based lane change intention recognition module was developed, employing a multilayer perceptron (MLP) to identify favored lane segments. By analyzing the relationship between these favored lane segments and the current lane occupied by the vehicle, the target vehicle's intentions to change lanes are inferred. Furthermore, by incorporating anchor points and fine-tuning vectors with lane change intentions, we predict the final positions of surrounding vehicles, thereby enhancing trajectory prediction accuracy based on these endpoint positions. Experimental results show that our proposed method surpasses existing target-driven technologies on the Argoverse 2 dataset, particularly in key performance metrics such as minADE, minFDE, and MR.