Path planning development for human-like virtual driver 2023-36-0068

Virtual simulation is a fundamental tool for the development of new vehicles, both for individual components and for complete subsystems and full vehicles. Many software tools exist in the automotive sector to assess full-vehicle behavior and performance, including multibody software and algorithms based on 14 (or more) degrees-of-freedom vehicle dynamics models. In order to reproduce the testing maneuvers and typical vehicle mission, a key part of such simulation tools is the virtual driver algorithm. It is essential to implement a control logic that reproduces the handling response of the driver, so that the closed-loop maneuvers can be evaluated. However, the response of typical virtual drivers is not always similar to the human driving characteristics. Virtual driver algorithms can perform very fast, precise, and smooth steering and pedal actions, while humans display a more variable, delayed and often not optimal actions.

The aim of this article is to describe the concept and implementation of a novel human-like path planning model. The algorithm is developed in MATLAB environment, creating a function that obtains a human-like path and vision logic by setting some key-parameters. They are: Distance Factor, Widening Factor, Cutting Factor, Inner Smoothing Factor and Outer Smoothing Factor.

The parameters - essential to alter the shape of the trajectory described in a track - have their values attributed by fitting experimental data gathered during test sessions in a driving simulator. The vehicle model used to implement the path planning system is based on the VI-Grade CarRealTime environment, in co-simulation with MATLAB/Simulink, and the results indicate that the novel algorithm has a closer correlation with the DiL tests than the original virtual driver. The stronger correlation is confirmed also in the comparison between different human drivers, showing that the proposed strategy is robust to driving styles.

Among the potential applications of this new “human-like virtual driver” approach is the ability to better predict human driver response during tuning and optimization of vehicles and control systems, apart from a further understanding of human driving behavior useful for tasks like ADAS and autonomous driving.