Enhancing vehicle trajectory prediction for V2V communication using a hybrid RNN approach

Abstract

This study presents a hybrid recurrent neural network (RNN) for predicting vehicle trajectories in terms of latitude and longitude positions, enabling V2 V communication. The hybrid network, which integrates Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) cells, is trained on the NGSIM dataset to address the regression problem of forecasting future vehicle positions. The proposed model achieves a root mean square error (RMSE) of <0.003, demonstrating a 33 % improvement compared to a network composed solely of LSTM cells. Furthermore, evaluation against recent approaches highlights the effectiveness of the proposed method in predicting vehicle trajectories. The impact of different dropout types and probabilities is also analyzed, with an input dropout probability of 0.6 delivering performance comparable to that of the model without dropout. These results indicate that the hybrid RNN effectively predicts future vehicle trajectories, laying a foundation for enhanced V2 V communication and contributing to advancements in autonomous vehicular systems.