Autonomous Vehicle Trajectory Combined Prediction model based on C-LSTM

There should be a complex driving environment formed by manned and unmanned vehicles with highly uncertain and dynamic interaction when autonomous vehicles enter actual traffic flow. Autonomous vehicles need to detect and analyze the movement of surrounding vehicles to make safe driving decisions. This paper proposes a Clustering Convolution-LSTM (CC-LSTM) vehicle trajectory prediction model which is made up two modules: Clustering module and Convolution-LSTM (C-LSTM) module. In order to find spatial-temporal features in vehicles trajectory, a clustering algorithm is proposed. Input data are constructed with temporal information by choosing the optimal history trajectory length. The convolution LSTM network layer extracts the spatial features of the trajectory and then make the trajectory prediction. Two modules are trained by different features data and combined. Simulation shows that the CC-LSTM prediction model with surrounding vehicle interaction information and selected features can meet the real-time and accuracy requirements of prediction.