Comparative Analysis of critical lane change for Autonomous Driving Vehicles and Human Drivers

In today’s industry, the need for trustworthy and safe high-driving automation algorithms for Autonomous Driving (AD) vehicles is rapidly growing. This growth is being driven by the expansion of the market and the enforcement of new safety regulations all around the world. To understand if and how an AD algorithm is improving the safety on the roads, the mainstream approach today is to test autonomous vehicles in critical scenarios modelled by real-life traffic data or natural trajectories databases. In this paper, we compare the behaviours of human drivers and AD vehicles when dealing with critical and ordinary scenarios of lane change manoeuvres on highways. The tested AD algorithm is based on the Apollo Open Autonomous Driving Platform (AOADP) which is integrated into the simulation. Additionally, we extract scenarios and human driving behaviours from the HighD dataset (a dataset of naturalistic vehicle trajectories recorded on German highways) and use the scenarios to test the AD algorithm. Eventually, the behaviours of the AD vehicles and human drivers are compared and analyzed. The simulation results reveal that the parameter configuration of the path planning algorithm has a significant impact on the execution of lane changes. The comparison between human driving behaviour and AD systems is one of the key factors in specifying the Operational Domain Design (ODD) of autonomous driving systems. The proposed approach opens a new door for quantitative analysis of the impact on autonomous driving measurements, thus contributing to a safe ODD specification.