Motion Planning Framework for Autonomous Vehicle with LSTM-based Predictive Model

Accurate motion prediction of surrounding vehicles is of great importance to enable lower risk motion planning for autonomous vehicles. In this paper, a long short-term memory-based motion prediction approach is first adopted to predict the trajectories of surrounding vehicles, and the designed Long-Short Term network is trained and tested based on the Next Generation Simulation (NGSIM) datasets. More importantly, a motion planning technique is developed based the trajectory prediction model. Specifically, to avoid collisions with surrounding vehicles on highway, an artificial potential field-based risk assessment method is presented, and the road boundaries are also considered. Subsequently, the model predictive control algorithm is utilized to consider the risk of collision, road boundaries as well as the vehicle dynamics. Finally, the simulation results show the effectiveness of the proposed method.