Vehicle Trajectory Prediction with Integrating a Physics based Method and a Data-based Method

Abstract

The physics and data-based methods are used to predict the trajectory of vehicles. To improve prediction performance, we suggest data-based methods using a deep learning model and a simple integration method. The integration method is the weighted sum, and the weights are extracted from the root mean square error of two methods. It shows enhanced results by taking the strength of both methods. The root mean square error of 0 to 3 seconds is less than 3 meter, and 3 to 6 seconds is less than 6 meter.