An Automated Lane-Change System Based on Probabilistic Trajectory Prediction Network

Abstract

In highway driving, understanding the intentions of surrounding vehicles is a crucial prerequisite to ensure collision-free lane changes. In this study, an automated lane change system framework is proposed for highway driving. A Long Short-Term Memory (LSTM)-based network is utilized to predict the paths of surrounding vehicles as probability distributions. When initiating a lane change, multiple candidate paths are generated, and the collision probability is then calculated by considering the generated paths of the host vehicle and the predicted paths of surrounding vehicles. Using the vehicle as a reference, the collision risk area is defined first related to the lane change. Secondly, the probability of the predicted distribution of the surrounding vehicles existing within this area is integrated to derive the collision probability. Subsequently, the collision-free optimal path is adopted, and Model Predictive Control (MPC) is employed for path tracking. The proposed framework was validated on a highway-like proving ground.