Driving intention prediction algorithm based on TPA-LSTM for autonomous vehicles

Abstract

To avoid the potential risk triggered by the failure of the conflict arbitration of autonomous vehicles, a driving intention prediction method based on the Long Short-Term Memory (LSTM) neural network involving Temporal Pattern Attention (TPA) is proposed. To be more specific, the TPA is embedded into the LSTM network to improve predictive accuracy. Furthermore, for evaluating the risk of the candidate trajectory, a risk assessment based on the velocity obstacle method which considers influence factors such as time to collision and collision energy loss is proposed. Finally, the proposed trajectory prediction algorithm is verified with the Next Generation Simulation data set and actual vehicle experiment. The results demonstrate the effectiveness of the proposed Method.