Trajectory and Velocity Prediction of Cut-In Vehicles With Deep Learning Method

Numerous studies have been conducted to predict lane-change trajectories. The significant differences between cut-ins and other lane changes suggest the necessity of building specialized algorithms tailored to learning vehicle cut-ins. In this paper, we explore predicting the trajectory and velocity of the cut-in vehicles with a deep learning method. Particularly, we propose a prediction algorithm by combining a Transformer-based encoder and an LSTM-based decoder. The Transformer-based encoder is applied to capture features related to the driving context of the cut-in vehicle. The LSTM decoder is employed to predict the trajectory and velocity of the cut-in vehicles by considering their temporal and social relationships. We extracted the cut-in events from NGSIM dataset for algorithm evaluation. We compared the performance of the proposed algorithm and three other deep learning algorithms based on the extracted cut-in events. The results suggest that the proposed algorithm outperforms other algorithms in trajectory and velocity predictions of the cut-in vehicles. Moreover, we analyze the effect of the historical data window size on the prediction performance of the proposed algorithm.