Risk potential estimation for autonomous vehicles based on risk-prediction of human driver in low-infrastructure city roads

Abstract

This paper proposes a control target generation algorithm for autonomous vehicles. The algorithm is applied to the risk potential method, in which some risks that human drivers are influenced by the surrounding environment information of the vehicle are defined by exponential functions. In addition, we proposed a method for estimating the parameters constituting the risk potential function in the algorithm. The proposed method uses driver's driving behavior data, vehicle status, and environmental information around the vehicle. Using this method, the possibility of traveling in urban areas where oncoming traffic, obstacles, and pedestrians exist on road traffic is also explained.