A novel behavior planning for human-like autonomous driving

With the development of intelligent vehicles for different types of drivers, generating different driving styles of driving schemes can improve the comfort of the automatic driving system. This paper proposes a human-like planning method for lane change and intersections. Firstly, in planning for the speed driving scheme to enter the intersection, a planning method is proposed to generate different types of speed schemes after determining the start and end speeds. Secondly, by analyzing the data collected under turning conditions on a driving simulator, we extracted the characteristics of driving behavior for each experiment and calculated the probabilistic statistical properties of turning radius and maximum lateral acceleration. Thirdly, the driving characteristics of lane changes are obtained through the fitting of the lane change dataset, and based on this, the probabilistic statistical properties of lane change behavior are determined. Fourthly, this paper proposes a method for deciding longitudinal acceleration by deep learning, by inputting the motion information of the surrounding traffic participants and the motion information of the host vehicle and outputting the decided acceleration, which has the characteristics of imitating the driving behaviors of the real drivers, and adding a safety limit to the output. Finally, simulations and tests are conducted on the VTD (Virtual Test Drive).