Supervised vehicle trajectory prediction using orthogonal image map for urban automated driving

In urban automated driving, it is important to generate appropriate behaviors according to surrounding circumstances. The automated vehicle is generally equipped with various sensors including LiDAR, Radar, Camera to percept the surrounding environment, estimate a precise location, optimize trajectories and control the vehicle. This study focuses on the trajectory generation for urban automated driving. The main approaches of trajectory generation are the mathematical model-based method and machine learning-based method. The former is able to guarantee time continuity of velocity and acceleration. On the other hand, machine learning-based approaches have been developed to learn the driving behavior of experienced drivers. This paper develops the method to predict vehicle future states as a vehicle trajectory based on supervised learning. The driving dataset is generated using actual sensor data and smooth trajectories which are created by the mathematical model. The relationship between the prediction accuracy and the input information was analyzed by evaluating different input information. The evaluated results show appropriate behavior according to the surrounding situations.