Overtaking Behavior Prediction of Rear Vehicle via LSTM Model

Predicting the driving behavior of surrounding vehicles is critical to perform motion planning tasks for autonomous vehicles. In this paper, we propose a dual long short-term memory (LSTM) framework to predict the overtaking behavior of the rear vehicle (RV). In the first layer of the framework, the softmax function is utilized to obtain the driving intention based on historical trajectory of RV. In the second layer of the framework, the encoder-decoder architecture is adopted to predict the future trajectory of RV. We trained and tested the proposed framework with the US-101 trajectory data of NGSIM dataset. Mean squared error (MSE) is used to evaluate the performance of the framework. The experimental results reveal that our framework can predict the overtaking intentions of RV 2–3 s before the overtaking maneuver occurs and conduct trajectory prediction at least 1.5 s earlier than that of single LSTM.